conference
stringclasses 6
values | title
stringlengths 2
176
| abstract
stringlengths 2
5k
| decision
stringclasses 11
values |
|---|---|---|---|
ICLR.cc/2023/Conference | ID and OOD Performance Are Sometimes Inversely Correlated on Real-world Datasets | Several studies have empirically compared in-distribution (ID) and out-of-distribution (OOD) performance of various models. They report frequent positive correlations on benchmarks in computer vision and NLP. Surprisingly, they never observe inverse correlations suggesting necessary trade-offs. This matters to determine whether ID performance can serve as a proxy for OOD generalization. This paper shows that inverse correlations between ID and OOD performance do happen in real-world benchmarks. They could be missed in past studies because of a biased selection of models. We show an example on the WILDS-Camelyon17 dataset, using models from multiple training epochs and random seeds. Our observations are particularly striking with models trained with a regularizer that diversifies the solutions to the ERM objective. We nuance recommendations and conclusions made in past studies. (1) High OOD performance may sometimes require trading off ID performance.(2) Focusing on ID performance alone may not lead to optimal OOD performance: it can lead to diminishing and eventually negative returns in OOD performance. (3) Our example reminds that empirical studies only chart regimes achievable with existing methods: care is warranted in deriving prescriptive recommendations. | Reject |
ICLR.cc/2022/Conference | Subspace Regularizers for Few-Shot Class Incremental Learning | Few-shot class incremental learning---the problem of updating a trained classifier to discriminate among an expanded set of classes with limited labeled data---is a key challenge for machine learning systems deployed in non-stationary environments. Existing approaches to the problem rely on complex model architectures and training procedures that are difficult to tune and re-use. In this paper, we present an extremely simple approach that enables the use of ordinary logistic regression classifiers for few-shot incremental learning. The key to this approach is a new family of \textit{subspace regularization} schemes that encourage weight vectors for new classes to lie close to the subspace spanned by the weights of existing classes. When combined with pretrained convolutional feature extractors, logistic regression models trained with subspace regularization outperform specialized, state-of-the-art approaches to few-shot incremental image classification by up to 23\% on the \textit{mini}ImageNet dataset. Because of its simplicity, subspace regularization can be straightforwardly configured to incorporate additional background information about the new classes (including class names and descriptions specified in natural language); this offers additional control over the trade-off between existing and new classes. Our results show that simple geometric regularization of class representations offers an effective tool for continual learning. | Accept (Poster) |
ICLR.cc/2020/Conference | Multichannel Generative Language Models | A channel corresponds to a viewpoint or transformation of an underlying meaning. A pair of parallel sentences in English and French express the same underlying meaning but through two separate channels corresponding to their languages. In this work, we present Multichannel Generative Language Models (MGLM), which models the joint distribution over multiple channels, and all its decompositions using a single neural network. MGLM can be trained by feeding it k way parallel-data, bilingual data, or monolingual data across pre-determined channels. MGLM is capable of both conditional generation and unconditional sampling. For conditional generation, the model is given a fully observed channel, and generates the k-1 channels in parallel. In the case of machine translation, this is akin to giving it one source, and the model generates k-1 targets. MGLM can also do partial conditional sampling, where the channels are seeded with prespecified words, and the model is asked to infill the rest. Finally, we can sample from MGLM unconditionally over all k channels. Our experiments on the Multi30K dataset containing English, French, Czech, and German languages suggest that the multitask training with the joint objective leads to improvements in bilingual translations. We provide a quantitative analysis of the quality-diversity trade-offs for different variants of the multichannel model for conditional generation, and a measurement of self-consistency during unconditional generation. We provide qualitative examples for parallel greedy decoding across languages and sampling from the joint distribution of the 4 languages. | Reject |
ICLR.cc/2020/Conference | UNIVERSAL MODAL EMBEDDING OF DYNAMICS IN VIDEOS AND ITS APPLICATIONS | Extracting underlying dynamics of objects in image sequences is one of the challenging problems in computer vision. On the other hand, dynamic mode decomposition (DMD) has recently attracted attention as a way of obtaining modal representations of nonlinear dynamics from (general multivariate time-series) data without explicit prior knowledge about the dynamics. In this paper, we propose a convolutional autoencoder based DMD (CAE-DMD) that is an extended DMD (EDMD) approach, to extract underlying dynamics in videos. To this end, we develop a modified CAE model by incorporating DMD on the encoder, which gives a more meaningful compressed representation of input image sequences. On the reconstruction side, a decoder is used to minimize the reconstruction error after applying the DMD, which in result gives an accurate reconstruction of inputs. We empirically investigated the performance of CAE-DMD in two applications: background/foreground extraction and video classification, on publicly available datasets. | Reject |
ICLR.cc/2019/Conference | GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding | For natural language understanding (NLU) technology to be maximally useful, it must be able to process language in a way that is not exclusive to a single task, genre, or dataset. In pursuit of this objective, we introduce the General Language Understanding Evaluation (GLUE) benchmark, a collection of tools for evaluating the performance of models across a diverse set of existing NLU tasks. By including tasks with limited training data, GLUE is designed to favor and encourage models that share general linguistic knowledge across tasks. GLUE also includes a hand-crafted diagnostic test suite that enables detailed linguistic analysis of models. We evaluate baselines based on current methods for transfer and representation learning and find that multi-task training on all tasks performs better than training a separate model per task. However, the low absolute performance of our best model indicates the need for improved general NLU systems. | Accept (Poster) |
ICLR.cc/2019/Conference | Predictive Uncertainty through Quantization | High-risk domains require reliable confidence estimates from predictive models.
Deep latent variable models provide these, but suffer from the rigid variational distributions used for tractable inference, which err on the side of overconfidence.
We propose Stochastic Quantized Activation Distributions (SQUAD), which imposes a flexible yet tractable distribution over discretized latent variables.
The proposed method is scalable, self-normalizing and sample efficient. We demonstrate that the model fully utilizes the flexible distribution, learns interesting non-linearities, and provides predictive uncertainty of competitive quality.
| Reject |
ICLR.cc/2021/Conference | Improved Uncertainty Post-Calibration via Rank Preserving Transforms | Modern machine learning models with high accuracy often exhibit poor uncertainty calibration: the output probabilities of the model do not reflect its accuracy, and tend to be over-confident. Existing post-calibration methods such as temperature scaling recalibrate a trained model using rather simple calibrators with one or few parameters, which can have a rather limited capacity. In this paper, we propose Neural Rank Preserving Transforms (NRPT), a new post-calibration method that adjusts the output probabilities of a trained classifier using a calibrator of higher capacity, while maintaining its prediction accuracy. NRPT learns a calibrator that preserves the rank of the probabilities through general monotonic transforms, individualizes to the original input, and allows learning with any loss function that encourages calibration. We show experimentally that NRPT improves the expected calibration error (ECE) significantly over existing post-calibration methods such as (local) temperature scaling on large-scale image and text classification tasks. The performance of NRPT can further match ensemble methods such as deep ensembles, while being much more parameter-efficient. We further demonstrate the improved calibration ability of NRPT beyond the ECE metric, such as accuracy among top-confidence predictions, as well as optimizing the tradeoff between calibration and sharpness. | Reject |
ICLR.cc/2023/Conference | Explaining Temporal Graph Models through an Explorer-Navigator Framework | While GNN explanation has recently received significant attention, existing works are consistently designed for static graphs. Due to the prevalence of temporal graphs, many temporal graph models have been proposed, but explaining their predictions remains to be explored. To bridge the gap, in this paper, we propose T-GNNExplainer for temporal graph model explanation. Specifically, we regard a temporal graph constituted by a sequence of temporal events. Given a target event, our task is to find a subset of previously occurred events that lead to the model's prediction for it. To handle this combinatorial optimization problem, T-GNNExplainer includes an explorer to find the event subsets with Monte Carlo Tree Search (MCTS) and a navigator that learns the correlations between events and helps reduce the search space. In particular, the navigator is trained in advance and then integrated with the explorer to speed up searching and achieve better results. To the best of our knowledge, T-GNNExplainer is the first explainer tailored for temporal graph models. We conduct extensive experiments to evaluate the performance of T-GNNExplainer. Experimental results on both real-world and synthetic datasets demonstrate that T-GNNExplainer can achieve superior performance with up to about 50% improvement in Area under Fidelity-Sparsity Curve. | Accept: poster |
ICLR.cc/2023/Conference | Exploring Visual Interpretability for Contrastive Language-Image Pretraining | Contrastive Language-Image Pre-training (CLIP) learns rich representations via readily available supervision of natural language. It improves the performance of downstream vision tasks, including but not limited to the zero-shot, long tail, segmentation, retrieval, caption, and video. However, the visual interpretability of CLIP is rarely studied, especially in the aspect of the raw feature map. To provide visual explanations of its predictions, we propose the Image-Text Similarity Map (ITSM). Based on it, we surprisingly find that CLIP prefers the background regions than the foregrounds, and shows erroneous visualization against human understanding. Experimentally, we find the devil is in the pooling part, where inappropriate pooling methods lead to a phenomenon called semantic shift. To correct and boost the visualization results, we propose the Masked Max Pooling, with attention map from the self-supervised image encoder. Meanwhile, interpretability and recognition require different representations. To address the problem, we propose the dual projections to cater this requirement. We integrate above methods as Interpretable Contrastive Language-Image Pre-training (ICLIP). Our experiments suggest that ICLIP greatly improves the interpretability of CLIP, e.g. nontrivial improvements at 32.85% and 49.10% on VOC 2012 dataset. | Withdrawn |
ICLR.cc/2019/Conference | Learning with Random Learning Rates. | Hyperparameter tuning is a bothersome step in the training of deep learning mod- els. One of the most sensitive hyperparameters is the learning rate of the gradient descent. We present the All Learning Rates At Once (Alrao) optimization method for neural networks: each unit or feature in the network gets its own learning rate sampled from a random distribution spanning several orders of magnitude. This comes at practically no computational cost. Perhaps surprisingly, stochastic gra- dient descent (SGD) with Alrao performs close to SGD with an optimally tuned learning rate, for various architectures and problems. Alrao could save time when testing deep learning models: a range of models could be quickly assessed with Alrao, and the most promising models could then be trained more extensively. This text comes with a PyTorch implementation of the method, which can be plugged on an existing PyTorch model. | Reject |
ICLR.cc/2022/Conference | Relational Learning with Variational Bayes | In psychology, relational learning refers to the ability to recognize and respond to relationship among objects irrespective of the nature of those objects. Relational learning has long been recognized as a hallmark of human cognition and a key question in artificial intelligence research. In this work, we propose an unsupervised learning method for addressing the relational learning problem where we learn the underlying relationship between a pair of data irrespective of the nature of those data. The central idea of the proposed method is to encapsulate the relational learning problem with a probabilistic graphical model in which we perform inference to learn about data relationship and other relational processing tasks. | Accept (Poster) |
ICLR.cc/2023/Conference | Phase transition for detecting a small community in a large network | How to detect a small community in a large network is an interesting problem, including clique detection as a special case, where a naive degree-based $\chi^2$-test was shown to be powerful in the presence of an Erdös-Renyi (ER) background. Using Sinkhorn's theorem, we show that the signal captured by the $\chi^2$-test may be a modeling artifact, and it may disappear once we replace the Erdös-Renyi model by a broader network model. We show that the recent SgnQ test is more appropriate for such a setting. The test is optimal in detecting communities with sizes comparable to the whole network, but has never been studied for our setting, which is substantially different and more challenging. Using a degree-corrected block model (DCBM), we establish phase transitions of this testing problem concerning the size of the small community and the edge densities in small and large communities. When the size of the small community is larger than $\sqrt{n}$, the SgnQ test is optimal for it attains the computational lower bound (CLB), the information lower bound for methods allowing polynomial computation time. When the size of the small community is smaller than $\sqrt{n}$, we establish the parameter regime where the SgnQ test has full power and make some conjectures of the CLB. We also study the classical information lower bound (LB) and show that there is always a gap between the CLB and LB in our range of interest. | Accept: poster |
ICLR.cc/2021/Conference | Non-greedy Gradient-based Hyperparameter Optimization Over Long Horizons | Gradient-based meta-learning has earned a widespread popularity in few-shot learning, but remains broadly impractical for tasks with long horizons (many gradient steps), due to memory scaling and gradient degradation issues. A common workaround is to learn meta-parameters online, but this introduces greediness which comes with a significant performance drop. In this work, we enable non-greedy meta-learning of hyperparameters over long horizons by sharing hyperparameters that are contiguous in time, and using the sign of hypergradients rather than their magnitude to indicate convergence. We implement this with forward-mode differentiation, which we extend to the popular momentum-based SGD optimizer. We demonstrate that the hyperparameters of this optimizer can be learned non-greedily without gradient degradation over $\sim 10^4$ inner gradient steps, by only requiring $\sim 10$ outer gradient steps. On CIFAR-10, we outperform greedy and random search methods for the same computational budget by nearly $10\%$. Code will be available upon publication. | Reject |
ICLR.cc/2023/Conference | Simple and Deep Graph Attention Networks | Graph Attention Networks (GATs) and Graph Convolutional Neural Networks (GCNs) are two state-of-the-art architectures in Graph Neural Networks (GNNs). It is well known that both models suffer from performance degradation when more GNN layers are stacked, and many works have been devoted to address this problem. We notice that main research efforts in the line focus on the GCN models, and their techniques cannot well fit the GATs models due to the inherent difference between these two architectures. In GATs, the attention mechanism ignores the overwhelming propagation from certain nodes as the number of layers increases. To sufficiently utilize the expressive power of GATs, we propose a new version of GAT named Layer-wise Self-adaptive GAT (LSGAT), which can effectively alleviate the oversmoothing issue in deep GATs. We redesign the attention coefficients computation mechanism adaptively adjusted by layer depth, which considers both immediate neighboring and non-adjacent nodes from a global view. The experimental evaluation confirms that LSGAT consistently achieves better results on node classification tasks over relevant counterparts. | Withdrawn |
ICLR.cc/2021/Conference | Optimizing Memory Placement using Evolutionary Graph Reinforcement Learning | For deep neural network accelerators, memory movement is both energetically expensive and can bound computation. Therefore, optimal mapping of tensors to memory hierarchies is critical to performance. The growing complexity of neural networks calls for automated memory mapping instead of manual heuristic approaches; yet the search space of neural network computational graphs have previously been prohibitively large. We introduce Evolutionary Graph Reinforcement Learning (EGRL), a method designed for large search spaces, that combines graph neural networks, reinforcement learning, and evolutionary search. A set of fast, stateless policies guide the evolutionary search to improve its sample-efficiency. We train and validate our approach directly on the Intel NNP-I chip for inference. EGRL outperforms policy-gradient, evolutionary search and dynamic programming baselines on BERT, ResNet-101 and ResNet-50. We additionally achieve 28-78% speed-up compared to the native NNP-I compiler on all three workloads. | Accept (Poster) |
ICLR.cc/2021/Conference | Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learning | In this work, we aim to develop a simple and scalable reinforcement learning algorithm that uses standard supervised learning methods as subroutines, while also being able to leverage off-policy data. Our proposed approach, which we refer to as advantage-weighted regression (AWR), consists of two standard supervised learning steps: one to regress onto target values for a value function, and another to regress onto weighted target actions for the policy. The method is simple and general, can accommodate continuous and discrete actions, and can be implemented in just a few lines of code on top of standard supervised learning methods. We provide a theoretical motivation for AWR and analyze its properties when incorporating off-policy data from experience replay. We evaluate AWR on a suite of standard OpenAI Gym benchmark tasks, and show that it achieves competitive performance compared to a number of well-established state-of-the-art RL algorithms. AWR is also able to acquire more effective policies than most off-policy algorithms when learning from purely static datasets with no additional environmental interactions. Furthermore, we demonstrate our algorithm on challenging continuous control tasks with highly complex simulated characters. | Reject |
ICLR.cc/2023/Conference | On the Lower Bound of Minimizing Polyak-Łojasiewicz functions | Polyak-Łojasiewicz (PL) [Polyak, 1963] condition is a weaker condition than the strong convexity but suffices to ensure a global convergence for the Gradient Descent algorithm. In this paper, we study the lower bound of algorithms using first-order oracles to find an approximate optimal solution. We show that any first-order algorithm requires at least $\Omega\left((L/\mu)^{1-\alpha} \right)$ gradient costs to find an $\epsilon$-approximate optimal solution for a general $L$-smooth function that has an $\mu$-PL constant for any $\alpha>0$. This result demonstrates the near optimality of the Gradient Descent algorithm to minimize smooth PL functions in the sense that there exists a ``hard'' PL function such that no first-order algorithm can be faster by a polynomial order. In contrast, it is well-known that the momentum technique, e.g. [Nesterov, 2003, chap. 2] can provably accelerate Gradient Descent to $O\left(\sqrt{L/\hat{\mu}}\log\frac{1}{\epsilon}\right)$ gradient costs for functions that are $L$-smooth and $\hat{\mu}$-strongly convex. Therefore, our result distinguishes the hardness of minimizing a smooth PL function and a smooth strongly convex function as the complexity of the former cannot be improved by any polynomial order in general. | Reject |
ICLR.cc/2022/Conference | Use of small auxiliary networks and scarce data to improve the adversarial robustness of deep learning models | Deep Learning models for image classification are known to be vulnerable to adversarial examples.
Adversarial training is one of the most effective ways to provide defense against such threats, however it is a cumbersome process which requires many data points and long computation times.
In a setting where only small amounts of data are available for this process, adversarial training may negatively impact the classification performance on clean images by overfitting on the small amount of data.
This would be undesirable, especially when a large pre-trained model with satisfactory performance on clean data is already available.
We propose a new strategy to make a previously-trained model more robust against adversarial attacks, using scarce data and without degrading its performance on clean samples.
The proposed strategy consists in freezing the parameters of the originally trained base model and adding small auxiliary networks along the architecture, which process the features to reduce the effect of any adversarial perturbation.
This method can be used to defend a model against any arbitrary attack.
A practical advantage of using auxiliary networks is that no modifications on the originally trained base model is required.
Therefore, it can serve as a patch or add on to fix large and expensive existing deep learning models with little additional resources.
Experiments on the CIFAR10 dataset showed that using only $10\%$ of the full training set, the proposed method was able to adequately defend the model against the AutoPGD attack while maintaining a classification accuracy on clean images outperforming the model with adversarial training by $7\%$. Indeed, the proposed method still performs reasonably well compared to adversarial training using $1\%$ of the full training set. | Withdrawn |
ICLR.cc/2023/Conference | EAGLE: Large-scale Learning of Turbulent Fluid Dynamics with Mesh Transformers | Estimating fluid dynamics is classically done through the simulation and integration of numerical models solving the Navier-Stokes equations, which is computationally complex and time-consuming even on high-end hardware. This is a notoriously hard problem to solve, which has recently been addressed with machine learning, in particular graph neural networks (GNN) and variants trained and evaluated on datasets of static objects in static scenes with fixed geometry. We attempt to go beyond existing work in complexity and introduce a new model, method and benchmark. We propose EAGLE: a large-scale dataset of ∼1.1 million 2D meshes resulting from simulations of unsteady fluid dynamics caused by a moving flow source interacting with nonlinear scene structure of varying geometries, with 600 different scenes of three different types in total. To perform future forecasting of pressure and velocity on the challenging EAGLE dataset, we introduce a new mesh transformer. It leverages node clustering, graph pooling and global attention to learn long-range dependencies between spatially distant data points without needing a large number of iterations, as existing GNN methods do. We show that our transformer outperforms state-of-the-art performance on, both, existing synthetic and real datasets and on EAGLE. Finally, we highlight that our approach learns to attend to airflow, integrating complex information in a single iteration. | Accept: poster |
ICLR.cc/2023/Conference | PromptCast: A New Prompt-based Learning Paradigm for Time Series Forecasting | This paper studies the time series forecasting problem from a whole new perspective. In the existing SOTA time-series representation learning methods, the forecasting models take a sequence of numerical values as input and yield numerical values as output. The existing SOTA models are largely based on Transformer architecture, modified with multiple encoding mechanisms to incorporate the context and semantics around the historical data. In this paper, we approach representation learning of time-series from the paradigm of prompt-based natural language modeling. Inspired by the successes of pre-trained language foundation models, we pose a question about whether these models can also be adapted to solve time-series forecasting. Thus, we propose a new forecasting paradigm: prompt-based time series forecasting (PromptCast). In this novel task, the numerical input and output are transformed into prompts. We frame the forecasting task in a sentence-to-sentence manner which makes it possible to directly apply language models for forecasting purposes. To support and facilitate the research of this task, we also present a large-scale dataset (PISA) that includes three real-world forecasting scenarios. We evaluate different SOTA numerical-based forecasting methods and language generation models such as Bart. The benchmark results with single- and multi-step forecasting settings demonstrate that the proposed prompt-based time series forecasting with language generation models is a promising research direction. In addition, in comparison to conventional numerical-based forecasting, PromptCast shows a much better generalization ability under the zero-shot setting. We believe that the proposed PromptCast task as well as our PISA dataset could provide novel insights and further lead to new research directions in the domain of time-series representation learning and forecasting. | Withdrawn |
ICLR.cc/2021/Conference | Revisiting Point Cloud Classification with a Simple and Effective Baseline | Processing point cloud data is an important component of many real-world systems. As such, a wide variety of point-based approaches have been proposed, reporting steady benchmark improvements over time. We study the key ingredients of this progress and uncover two critical results. First, we find that auxiliary factors like different evaluation schemes, data augmentation strategies, and loss functions, which are independent of the model architecture, make a large difference in performance. The differences are large enough that they obscure the effect of architecture. When these factors are controlled for, PointNet++, a relatively older network, performs competitively with recent methods. Second, a very simple projection-based method, which we refer to as SimpleView, performs surprisingly well. It achieves on par or better results than sophisticated state-of-the-art methods on ModelNet40, while being half the size of PointNet++. It also outperforms state-of-the-art methods on ScanObjectNN, a real-world point cloud benchmark, and demonstrates better cross-dataset generalization.
| Reject |
ICLR.cc/2022/Conference | ADAVI: Automatic Dual Amortized Variational Inference Applied To Pyramidal Bayesian Models | Frequently, population studies feature pyramidally-organized data represented using Hierarchical Bayesian Models (HBM) enriched with plates. These models can become prohibitively large in settings such as neuroimaging, where a sample is composed of a functional MRI signal measured on 300 brain locations, across 4 measurement sessions, and 30 subjects, resulting in around 1 million latent parameters.
Such high dimensionality hampers the usage of modern, expressive flow-based techniques.
To infer parameter posterior distributions in this challenging class of problems, we designed a novel methodology that automatically produces a variational family dual to a target HBM. This variational family, represented as a neural network, consists in the combination of an attention-based hierarchical encoder feeding summary statistics to a set of normalizing flows. Our automatically-derived neural network exploits exchangeability in the plate-enriched HBM and factorizes its parameter space. The resulting architecture reduces by orders of magnitude its parameterization with respect to that of a typical flow-based representation, while maintaining expressivity.
Our method performs inference on the specified HBM in an amortized setup: once trained, it can readily be applied to a new data sample to compute the parameters' full posterior.
We demonstrate the capability and scalability of our method on simulated data, as well as a challenging high-dimensional brain parcellation experiment. We also open up several questions that lie at the intersection between normalizing flows, SBI, structured Variational Inference, and inference amortization. | Accept (Poster) |
ICLR.cc/2020/Conference | Projection-Based Constrained Policy Optimization | We consider the problem of learning control policies that optimize a reward function while satisfying constraints due to considerations of safety, fairness, or other costs. We propose a new algorithm - Projection-Based Constrained Policy Optimization (PCPO), an iterative method for optimizing policies in a two-step process - the first step performs an unconstrained update while the second step reconciles the constraint violation by projecting the policy back onto the constraint set. We theoretically analyze PCPO and provide a lower bound on reward improvement, as well as an upper bound on constraint violation for each policy update. We further characterize the convergence of PCPO with projection based on two different metrics - L2 norm and Kullback-Leibler divergence. Our empirical results over several control tasks demonstrate that our algorithm achieves superior performance, averaging more than 3.5 times less constraint violation and around 15% higher reward compared to state-of-the-art methods. | Accept (Poster) |
ICLR.cc/2023/Conference | Lossy Compression with Gaussian Diffusion | We consider a novel lossy compression approach based on unconditional diffusion generative models, which we call DiffC. Unlike modern compression schemes which rely on transform coding and quantization to restrict the transmitted information, DiffC relies on the efficient communication of pixels corrupted by Gaussian noise. We implement a proof of concept and find that it works surprisingly well despite the lack of an encoder transform, outperforming the state-of-the-art generative compression method HiFiC on ImageNet 64x64. DiffC only uses a single model to encode and denoise corrupted pixels at arbitrary bitrates. The approach further provides support for progressive coding, that is, decoding from partial bit streams. We perform a rate-distortion analysis to gain a deeper understanding of its performance, providing analytical results for multivariate Gaussian data as well as theoretic bounds for general distributions. Furthermore, we prove that a flow-based reconstruction achieves a 3 dB gain over ancestral sampling at high bitrates. | Reject |
ICLR.cc/2022/Conference | Evaluating generative networks using Gaussian mixtures of image features | We develop a measure for evaluating the performance of generative networks given two sets of images. A popular performance measure currently used to do this is the Fréchet Inception Distance (FID). However, FID assumes that images featurized using the penultimate layer of Inception-v3 follow a Gaussian distribution. This assumption allows FID to be easily computed, since FID uses the 2-Wasserstein distance of two Gaussian distributions fitted to the featurized images. However, we show that Inception-v3 features of the ImageNet dataset are not Gaussian; in particular, each marginal is not Gaussian. To remedy this problem, we model the featurized images using Gaussian mixture models (GMMs) and compute the $2$-Wasserstein distance restricted to GMMs. We define a performance measure, which we call WaM, on two sets of images by using Inception-v3 (or another classifier) to featurize the images, estimate two GMMs, and use the restricted 2-Wasserstein distance to compare the GMMs. We experimentally show the advantages of WaM over FID, including how FID is more sensitive than WaM to image perturbations. By modelling the non-Gaussian features obtained from Inception-v3 as GMMs and using a GMM metric, we can more accurately evaluate generative network performance. | Reject |
ICLR.cc/2021/Conference | Optimizing Large-Scale Hyperparameters via Automated Learning Algorithm | Modern machine learning algorithms usually involve tuning multiple (from one to thousands) hyperparameters which play a pivotal role in terms of model generalizability. Globally choosing appropriate values of hyperparameters is extremely computationally challenging. Black-box optimization and gradient-based algorithms are two dominant approaches to hyperparameter optimization while they have totally distinct advantages. How to design a new hyperparameter optimization technique inheriting all benefits from both approaches is still an open problem. To address this challenging problem, in this paper, we propose a new hyperparameter optimization method with zeroth-order hyper-gradients (HOZOG). Specifically, we first exactly formulate hyperparameter optimization as an $\mathcal{A}$-based constrained optimization problem, where $\mathcal{A}$ is a black-box optimization algorithm (such as deep neural network). Then, we use the average zeroth-order hyper-gradients to update hyperparameters. We provide the feasibility analysis of using HOZOG to achieve hyperparameter optimization. The experimental results on three representative hyperparameter (the size is from 1 to 1250) optimization tasks demonstrate the benefits of HOZOG in terms of \textit{simplicity, scalability, flexibility, effectiveness and efficiency} compared with the state-of-the-art hyperparameter optimization methods. | Reject |
ICLR.cc/2021/Conference | Searching for Convolutions and a More Ambitious NAS | An important goal of neural architecture search (NAS) is to automate-away the design of neural networks on new tasks in under-explored domains, thus helping to democratize machine learning. However, current NAS research largely focuses on search spaces consisting of existing operations---such as different types of convolution---that are already known to work well on well-studied problems---often in computer vision. Our work is motivated by the following question: can we enable users to build their own search spaces and discover the right neural operations given data from their specific domain? We make progress towards this broader vision for NAS by introducing a space of operations generalizing the convolution that enables search over a large family of parameterizable linear-time matrix-vector functions. Our flexible construction allows users to design their own search spaces adapted to the nature and shape of their data, to warm-start search methods using convolutions when they are known to perform well, or to discover new operations from scratch when they do not. We evaluate our approach on several novel search spaces over vision and text data, on all of which simple NAS search algorithms can find operations that perform better than baseline layers. | Reject |
ICLR.cc/2022/Conference | Expressivity of Emergent Languages is a Trade-off between Contextual Complexity and Unpredictability | Researchers are using deep learning models to explore the emergence of language in various language games, where agents interact and develop an emergent language to solve tasks. We focus on the factors that determine the expressivity of emergent languages, which reflects the amount of information about input spaces those languages are capable of encoding. We measure the expressivity of emergent languages based on the generalisation performance across different games, and demonstrate that the expressivity of emergent languages is a trade-off between the complexity and unpredictability of the context those languages emerged from. Another contribution of this work is the discovery of message type collapse, i.e. the number of unique messages is lower than that of inputs. We also show that using the contrastive loss proposed by Chen et al. (2020) can alleviate this problem. | Accept (Poster) |
ICLR.cc/2023/Conference | Solving stochastic weak Minty variational inequalities without increasing batch size | This paper introduces a family of stochastic extragradient-type algorithms for a class of nonconvex-nonconcave problems characterized by the weak Minty variational inequality (MVI). Unlike existing results on extragradient methods in the monotone setting, employing diminishing stepsizes is no longer possible in the weak MVI setting. This has led to approaches such as increasing batch sizes per iteration which can however be prohibitively expensive. In contrast, our proposed methods involves two stepsizes and only requires one additional oracle evaluation per iteration. We show that it is possible to keep one fixed stepsize while it is only the second stepsize that is taken to be diminishing, making it interesting even in the monotone setting. Almost sure convergence is established and we provide a unified analysis for this family of schemes which contains a nonlinear generalization of the celebrated primal dual hybrid gradient algorithm.
| Accept: poster |
ICLR.cc/2020/Conference | Learning to Learn via Gradient Component Corrections | Gradient-based meta-learning algorithms require several steps of gradient descent to adapt to newly incoming tasks. This process becomes more costly as the number of samples increases. Moreover, the gradient updates suffer from several sources of noise leading to a degraded performance. In this work, we propose a meta-learning algorithm equipped with the GradiEnt Component COrrections, aGECCO cell for short, which generates a multiplicative corrective low-rank matrix which (after vectorization) corrects the estimated gradients. GECCO contains a simple decoder-like network with learnable parameters, an attention module and a so-called context input parameter. The context parameter of GECCO is updated to generate a low-rank corrective term for the network gradients. As a result, meta-learning requires only a few of gradient updates to absorb new task (often, a single update is sufficient in the few-shot scenario). While previous approaches address this problem by altering the learning rates, factorising network parameters or directly learning feature corrections from features and/or gradients, GECCO is an off-the-shelf generator-like unit that performs element-wise gradient corrections without the need to ‘observe’ the features and/or the gradients directly. We show that our GECCO (i) accelerates learning, (ii) performs robust corrections of the gradients corrupted by a noise, and (iii) leads to notable improvements over existing gradient-based meta-learning algorithms. | Withdrawn |
ICLR.cc/2023/Conference | Trimsformer: Trimming Transformer via Searching for Low-Rank Structure | Vision Transformers (ViT) have recently been used successfully in various computer vision tasks, but the high computational cost hinders their practical deployment. One of the most well-known methods to alleviate computational burden is low-rank approximation. However, how to automatically search for a low-rank configuration efficiently remains a challenge. In this paper, we propose Trimsformer, an end-to-end automatic low-rank approximation framework based on a neural architecture search scheme, which tackles the inefficiency of searching for a target low-rank configuration out of numerous ones. We propose weight inheritance which encodes enormous rank choices into a single search space. In addition, we share the gradient information among building blocks to boost the convergence of the supernet training. Furthermore, to mitigate the initial performance gap between subnetworks caused by using pre-trained weights, we adopt non-uniform sampling to promote the overall subnetwork performance. Extensive results show the efficacy of our Trimsformer framework. For instance, with our method, Trim-DeiT-B/Trim-Swin-B can save up to 57%/46% FLOPs with 1.1%/0.2% higher accuracy over DeiT-B/Swin-B. Last but not least, Trimsformer exhibits remarkable generality and orthogonality. We can yield extra 21%$\sim$26% FLOPs reductions on top of the popular compression method as well as the compact hybrid structure. Our code will be released. | Withdrawn |
ICLR.cc/2021/Conference | Action and Perception as Divergence Minimization | We introduce a unified objective for action and perception of intelligent agents. Extending representation learning and control, we minimize the joint divergence between the combined system of agent and environment and a target distribution. Intuitively, such agents use perception to align their beliefs with the world, and use actions to align the world with their beliefs. Minimizing the joint divergence to an expressive target maximizes the mutual information between the agent's representations and inputs, thus inferring representations that are informative of past inputs and exploring future inputs that are informative of the representations. This lets us explain intrinsic objectives, such as representation learning, information gain, empowerment, and skill discovery from minimal assumptions. Moreover, interpreting the target distribution as a latent variable model suggests powerful world models as a path toward highly adaptive agents that seek large niches in their environments, rendering task rewards optional. The framework provides a common language for comparing a wide range of objectives, advances the understanding of latent variables for decision making, and offers a recipe for designing novel objectives. We recommend deriving future agent objectives the joint divergence to facilitate comparison, to point out the agent's target distribution, and to identify the intrinsic objective terms needed to reach that distribution. | Reject |
ICLR.cc/2021/Conference | How Important is Importance Sampling for Deep Budgeted Training? | Long iterative training processes for Deep Neural Networks (DNNs) are commonly required to achieve state-of-the-art performance in many computer vision tasks. Core-set selection and importance sampling approaches might play a key role in budgeted training regimes, i.e.~when limiting the number of training iterations. The former demonstrate that retaining informative samples is important to avoid large drops in accuracy, and the later aim at dynamically estimating the sample importance to speed-up convergence. This work explores this paradigm and how a budget constraint interacts with importance sampling approaches and data augmentation techniques. We show that under budget restrictions, importance sampling approaches do not provide a consistent improvement over uniform sampling. We suggest that, given a specific budget, the best course of action is to disregard the importance and introduce adequate data augmentation. For example, training in CIFAR-10/100 with 30% of the full training budget, a uniform sampling strategy with certain data augmentation surpasses the performance of 100% budget models trained with standard data augmentation. We conclude from our work that DNNs under budget restrictions benefit greatly from variety in the samples and that finding the right samples to train is not the most effective strategy when balancing high performance with low computational requirements. The code will be released after the review process. | Reject |
ICLR.cc/2023/Conference | Adapting Pre-trained Language Models for Quantum Natural Language Processing | The emerging classical-quantum transfer learning paradigm has brought a decent performance to quantum computational models in many tasks, such as computer vision, by enabling a combination of quantum models and classical pre-trained neural networks. However, using quantum computing with pre-trained models has yet been explored in natural language processing (NLP). Due to the high linearity constraints of the underlying quantum computing infrastructures, existing Quantum NLP models are limited in performance on real tasks. We fill this gap by pre-training a sentence state with complex-valued BERT-like architecture, and adapting it to the classical-quantum transfer learning scheme for sentence classification. On quantum simulation experiments, the pre-trained representation can bring 50% to 60% increases to the capacity of end-to-end quantum models. | Withdrawn |
ICLR.cc/2022/Conference | On Deep Neural Network Calibration by Regularization and its Impact on Refinement | Deep neural networks have been shown to be highly miscalibrated. often they tend to be overconfident in their predictions. It poses a significant challenge for safety-critical systems to utilise deep neural networks (DNNs), reliably. Many recently proposed approaches to mitigate this have demonstrated substantial progress in improving DNN calibration. However, they hardly touch upon refinement, which historically has been an essential aspect of calibration. Refinement indicates separability of a network's correct and incorrect predictions. This paper presents a theoretically and empirically supported exposition reviewing refinement of a calibrated model. Firstly, we show the breakdown of expected calibration error (ECE), into predicted confidence and refinement under the assumption of over-confident predictions. Secondly, linking with this result, we highlight that regularisation based calibration only focuses on naively reducing a model's confidence. This logically has a severe downside to a model's refinement as correct and incorrect predictions become tightly coupled. Lastly, connecting refinement with ECE also provides support to existing refinement based approaches which improve calibration but do not explain the reasoning behind it. We support our claims through rigorous empirical evaluations of many state of the art calibration approaches on widely used datasets and neural networks. We find that many calibration approaches with the likes of label smoothing, mixup etc. lower the usefulness of a DNN by degrading its refinement. Even under natural data shift, this calibration-refinement trade-off holds for the majority of calibration methods. | Withdrawn |
ICLR.cc/2020/Conference | Deep Imitative Models for Flexible Inference, Planning, and Control | Imitation Learning (IL) is an appealing approach to learn desirable autonomous behavior. However, directing IL to achieve arbitrary goals is difficult. In contrast, planning-based algorithms use dynamics models and reward functions to achieve goals. Yet, reward functions that evoke desirable behavior are often difficult to specify. In this paper, we propose "Imitative Models" to combine the benefits of IL and goal-directed planning. Imitative Models are probabilistic predictive models of desirable behavior able to plan interpretable expert-like trajectories to achieve specified goals. We derive families of flexible goal objectives, including constrained goal regions, unconstrained goal sets, and energy-based goals. We show that our method can use these objectives to successfully direct behavior. Our method substantially outperforms six IL approaches and a planning-based approach in a dynamic simulated autonomous driving task, and is efficiently learned from expert demonstrations without online data collection. We also show our approach is robust to poorly-specified goals, such as goals on the wrong side of the road. | Accept (Poster) |
ICLR.cc/2020/Conference | Enhancing Attention with Explicit Phrasal Alignments | The attention mechanism is an indispensable component of any state-of-the-art neural machine translation system. However, existing attention methods are often token-based and ignore the importance of phrasal alignments, which are the backbone of phrase-based statistical machine translation. We propose a novel phrase-based attention method to model n-grams of tokens as the basic attention entities, and design multi-headed phrasal attentions within the Transformer architecture to perform token-to-token and token-to-phrase mappings. Our approach yields improvements in English-German, English-Russian and English-French translation tasks on the standard WMT'14 test set. Furthermore, our phrasal attention method shows improvements on the one-billion-word language modeling benchmark.
| Reject |
ICLR.cc/2023/Conference | Making Better Decision by Directly Planning in Continuous Control | By properly utilizing the learned environment model, model-based reinforcement learning methods can improve the sample efficiency for decision-making problems. Beyond using the learned environment model to train a policy, the success of MCTS-based methods shows that directly incorporating the learned environment model as a planner to make decisions might be more effective. However, when action space is of high dimension and continuous, directly planning according to the learned model is costly and non-trivial. Because of two challenges: (1) the infinite number of candidate actions and (2) the temporal dependency between actions in different timesteps. To address these challenges, inspired by Differential Dynamic Programming (DDP) in optimal control theory, we design a novel Policy Optimization with Model Planning (POMP) algorithm, which incorporates a carefully designed Deep Differential Dynamic Programming (D3P) planner into the model-based RL framework. In D3P planner, (1) to effectively plan in the continuous action space, we construct a locally quadratic programming problem that uses a gradient-based optimization process to replace search. (2) To take the temporal dependency of actions at different timesteps into account, we leverage the updated and latest actions of previous timesteps (i.e., step $1, \cdots, h-1$) to update the action of the current step (i.e., step $h$), instead of updating all actions simultaneously. We theoretically prove the convergence rate for our D3P planner and analyze the effect of the feedback term. In practice, to effectively apply the neural network based D3P planner in reinforcement learning, we leverage the policy network to initialize the action sequence and keep the action update conservative in the planning process. Experiments demonstrate that POMP consistently improves sample efficiency on widely used continuous control tasks. Our code is released at https://github.com/POMP-D3P/POMP-D3P. | Accept: poster |
ICLR.cc/2023/Conference | Targeted Hyperparameter Optimization with Lexicographic Preferences Over Multiple Objectives | Motivated by various practical applications, we propose a novel and general formulation of targeted multi-objective hyperparameter optimization. Our formulation allows a clear specification of an automatable optimization goal using lexicographic preference over multiple objectives. We then propose a randomized directed search method named LexiFlow to solve this problem. We demonstrate the strong empirical performance of the proposed algorithm in multiple hyperparameter optimization tasks. | Accept: notable-top-5% |
ICLR.cc/2019/Conference | Pyramid Recurrent Neural Networks for Multi-Scale Change-Point Detection | Many real-world time series, such as in activity recognition, finance, or climate science, have changepoints where the system's structure or parameters change. Detecting changes is important as they may indicate critical events. However, existing methods for changepoint detection face challenges when (1) the patterns of change cannot be modeled using simple and predefined metrics, and (2) changes can occur gradually, at multiple time-scales. To address this, we show how changepoint detection can be treated as a supervised learning problem, and propose a new deep neural network architecture that can efficiently identify both abrupt and gradual changes at multiple scales. Our proposed method, pyramid recurrent neural network (PRNN), is designed to be scale-invariant, by incorporating wavelets and pyramid analysis techniques from multi-scale signal processing. Through experiments on synthetic and real-world datasets, we show that PRNN can detect abrupt and gradual changes with higher accuracy than the state of the art and can extrapolate to detect changepoints at novel timescales that have not been seen in training. | Reject |
ICLR.cc/2022/Conference | On the relationship between disentanglement and multi-task learning | One of the main arguments behind studying disentangled representations is the assumption that they can be easily reused in different tasks. At the same time finding a joint, adaptable representation of data is one of the key challenges in the multi-task learning setting. In this paper, we take a closer look at the relationship between disentanglement and multi-task learning based on hard parameter sharing. We perform a thorough empirical study of the representations obtained by neural networks trained on automatically generated supervised tasks. Using a set of standard metrics we show that disentanglement appears in a natural way during the process of multi-task neural network training. | Reject |
ICLR.cc/2022/Conference | Differentiable Expectation-Maximization for Set Representation Learning | We tackle the set2vec problem, the task of extracting a vector representation from an input set comprised of a variable number of feature vectors. Although recent approaches based on self attention such as (Set)Transformers were very successful due to the capability of capturing complex interaction between set elements, the computational overhead is the well-known downside. The inducing-point attention and the latest optimal transport kernel embedding (OTKE) are promising remedies that attain comparable or better performance with reduced computational cost, by incorporating a fixed number of learnable queries in attention. In this paper we approach the set2vec problem from a completely different perspective. The elements of an input set are considered as i.i.d.~samples from a mixture distribution, and we define our set embedding feed-forward network as the maximum-a-posterior (MAP) estimate of the mixture which is approximately attained by a few Expectation-Maximization (EM) steps. The whole MAP-EM steps are differentiable operations with a fixed number of mixture parameters, allowing efficient auto-diff back-propagation for any given downstream task. Furthermore, the proposed mixture set data fitting framework allows unsupervised set representation learning naturally via marginal likelihood maximization aka the empirical Bayes. Interestingly, we also find that OTKE can be seen as a special case of our framework, specifically a single-step EM with extra balanced assignment constraints on the E-step. Compared to OTKE, our approach provides more flexible set embedding as well as prior-induced model regularization. We evaluate our approach on various tasks demonstrating improved performance over the state-of-the-arts. | Accept (Poster) |
ICLR.cc/2019/Conference | Successor Uncertainties: exploration and uncertainty in temporal difference learning | We consider the problem of balancing exploration and exploitation in sequential decision making problems. This trade-off naturally lends itself to probabilistic modelling. For a probabilistic approach to be effective, considering uncertainty about all immediate and long-term consequences of agent's actions is vital. An estimate of such uncertainty can be leveraged to guide exploration even in situations where the agent needs to perform a potentially long sequence of actions before reaching an under-explored area of the environment. This observation was made by the authors of the Uncertainty Bellman Equation model (O'Donoghue et al., 2018), which explicitly considers full marginal uncertainty for each decision the agent faces. However, their model still considers a fully factorised posterior over the consequences of each action, meaning that dependencies vital for correlated long-term exploration are ignored. We go a step beyond and develop Successor Uncertainties, a probabilistic model for the state-action value function of a Markov Decision Process with a non-factorised covariance. We demonstrate how this leads to greatly improved performance on classic tabular exploration benchmarks and show strong performance of our method on a subset of Atari baselines. Overall, Successor Uncertainties provides a better probabilistic model for temporal difference learning at a similar computational cost to its predecessors. | Reject |
ICLR.cc/2021/Conference | cross-modal knowledge enhancement mechanism for few-shot learning | Few-shot learning problems require models to recognize novel classes with only a few supported samples. However, it remains challenging for the model to generalize novel classes with such limited samples. Driven by human behavior, researchers introduced semantic information (e.g. novel categories descriptions, label names, etc.) onto existing methods as prior knowledge to generalize more precise class representations. Despite the promising performance, these methods are under the assumption that users are able to provide precise semantic information for all target categories and this is hard to be satisfied in a real scenario. To address this problem, we proposed a novel Cross-modality Knowledge Enhancement Mechanism(CKEM) to discover task-relevant information in external semantic knowledge automatically. CKEM first utilizes Cross-modality Graph Builder(CGB) to align two unitary modality information (support labeled images and external semantic knowledge) into a cross-modality knowledge graph. After that, with the message-passing mechanism, CKEM selects and transfers relevant knowledge from external semantic knowledge bank to original visual-based class representations in Knowledge Fusion Model(KFM). Through a series of experiments, we show that our method improves the existing metric-based meta-learning methods with 1\% - 5\% for 1-shot and 5-shot settings on both mini-ImageNet and tiered-ImageNet datasets. | Withdrawn |
ICLR.cc/2021/Conference | Rethinking Sampling in 3D Point Cloud Generative Adversarial Networks | In this paper, we examine the long-neglected yet important effects of point sam- pling patterns in point cloud GANs. Through extensive experiments, we show that sampling-insensitive discriminators (e.g. PointNet-Max) produce shape point clouds with point clustering artifacts while sampling-oversensitive discriminators (e.g. PointNet++, DGCNN, PointConv, KPConv) fail to guide valid shape generation. We propose the concept of sampling spectrum to depict the different sampling sensitivities of discriminators. We further study how different evaluation metrics weigh the sampling pattern against the geometry and propose several perceptual metrics forming a sampling spectrum of metrics. Guided by the proposed sampling spectrum, we discover a middle-point sampling-aware baseline discriminator, PointNet-Mix, which improves all existing point cloud generators by a large margin on sampling-related metrics. We point out that, given that recent research has been focused on the generator design, the discriminator design needs more attention. Our work provides both suggestions and tools for building future discriminators. We will release the code to facilitate future research. | Reject |
ICLR.cc/2023/Conference | Annealed Fisher Implicit Sampler | Sampling from an un-normalized target distribution is an important problem in many scientific fields. An implicit sampler uses a parametric transform $x=G_\theta(z)$ to push forward an easy-to-sample latent code $z$ to obtain a sample $x$. Such samplers are favored for fast inference speed and flexible architecture. Thus it is appealing to train an implicit sampler for sampling from the un-normalized target. In this paper, we propose a novel approach to training an implicit sampler by minimizing the Fisher Divergence between sampler and target distribution. We find that the trained sampler works well for relatively simple targets but may fail for more complicated multi-modal targets. To improve the training for multi-modal targets, we propose another adaptive training approach that trains the sampler to gradually learn a sequence of annealed distributions. We construct the annealed distribution path to bridge a simple distribution and the complicated target. With the annealed approach, the sampler is capable of handling challenging multi-modal targets. In addition, we also introduce a few MCMC correction steps after the sampler to better spread the samples. We call our proposed sampler \emph{the Annealed Fisher Implicit Sampler} (AFIS). We test AFIS on several sampling benchmarks. The experiments show that our AFIS outperforms baseline methods in many aspects. We also show in theory that the added MC correction steps get faster mixing by using the learned sampler as MCMC's initialization.
| Reject |
ICLR.cc/2019/Conference | Padam: Closing the Generalization Gap of Adaptive Gradient Methods in Training Deep Neural Networks | Adaptive gradient methods, which adopt historical gradient information to automatically adjust the learning rate, despite the nice property of fast convergence, have been observed to generalize worse than stochastic gradient descent (SGD) with momentum in training deep neural networks. This leaves how to close the generalization gap of adaptive gradient methods an open problem. In this work, we show that adaptive gradient methods such as Adam, Amsgrad, are sometimes "over adapted". We design a new algorithm, called Partially adaptive momentum estimation method (Padam), which unifies the Adam/Amsgrad with SGD by introducing a partial adaptive parameter p, to achieve the best from both worlds. Experiments on standard benchmarks show that Padam can maintain fast convergence rate as Adam/Amsgrad while generalizing as well as SGD in training deep neural networks. These results would suggest practitioners pick up adaptive gradient methods once again for faster training of deep neural networks. | Reject |
ICLR.cc/2019/Conference | Incremental Hierarchical Reinforcement Learning with Multitask LMDPs | Exploration is a well known challenge in Reinforcement Learning. One principled way of overcoming this challenge is to find a hierarchical abstraction of the base problem and explore at these higher levels, rather than in the space of primitives. However, discovering a deep abstraction autonomously remains a largely unsolved problem, with practitioners typically hand-crafting these hierarchical control architectures. Recent work with multitask linear Markov decision processes, allows for the autonomous discovery of deep hierarchical abstractions, but operates exclusively in the offline setting. By extending this work, we develop an agent that is capable of incrementally growing a hierarchical representation, and using its experience to date to improve exploration. | Reject |
ICLR.cc/2022/Conference | Large Language Models Can Be Strong Differentially Private Learners | Differentially Private (DP) learning has seen limited success for building large deep learning models of text, and straightforward attempts at applying Differentially Private Stochastic Gradient Descent (DP-SGD) to NLP tasks have resulted in large performance drops and high computational overhead.
We show that this performance drop can be mitigated with (1) the use of large pretrained language models; (2) non-standard hyperparameters that suit DP optimization; and (3) fine-tuning objectives which are aligned with the pretraining procedure.
With the above, we obtain NLP models that outperform state-of-the-art DP-trained models under the same privacy budget and strong non-private baselines---by directly fine-tuning pretrained models with DP optimization on moderately-sized corpora.
To address the computational challenge of running DP-SGD with large Transformers, we propose a memory saving technique that allows clipping in DP-SGD to run without instantiating per-example gradients for any linear layer in the model.
The technique enables privately training Transformers with almost the same memory cost as non-private training at a modest run-time overhead.
Contrary to conventional wisdom that DP optimization fails at learning high-dimensional models (due to noise that scales with dimension) empirical results reveal that private learning with pretrained language models tends to not suffer from dimension-dependent performance degradation.
Code to reproduce results can be found at https://github.com/lxuechen/private-transformers.
| Accept (Oral) |
ICLR.cc/2023/Conference | Dynamics-aware Skill Generation from Behaviourally Diverse Demonstrations | Learning from demonstrations (LfD) provides a data-efficient way for a robot to learn a task by observing humans performing the task, without the need for an explicit reward function. However, in many real-world scenarios (e.g., driving a car) humans often perform the same task in different ways, motivated not only by the primary objective of the task (e.g., reaching the destination safely) but also by their individual preferences (e.g., different driving behaviours), leading to a multi-modal distribution of demonstrations. In this work, we consider an LfD problem, where the reward function for the main objective of the task is known to the learning agent; however, the individual preferences leading to the variations in the demonstrations are unknown. We show that current LfD approaches learn policies that either track a single mode or the mean of the demonstration distribution. In contrast, we propose an algorithm to learn a diverse set of policies to perform the task, capturing the different modes in the demonstrations due to the diverse preferences of the individuals. We show that we can build a parameterised solution space that captures different behaviour patterns from the demonstrations. Then, a set of policies can be generated in solution space that generate a diverse range of behaviours that go beyond the provided demonstrations. | Reject |
ICLR.cc/2021/Conference | Meta-Learning of Structured Task Distributions in Humans and Machines | In recent years, meta-learning, in which a model is trained on a family of tasks (i.e. a task distribution), has emerged as an approach to training neural networks to perform tasks that were previously assumed to require structured representations, making strides toward closing the gap between humans and machines. However, we argue that evaluating meta-learning remains a challenge, and can miss whether meta-learning actually uses the structure embedded within the tasks. These meta-learners might therefore still be significantly different from humans learners. To demonstrate this difference, we first define a new meta-reinforcement learning task in which a structured task distribution is generated using a compositional grammar. We then introduce a novel approach to constructing a "null task distribution" with the same statistical complexity as this structured task distribution but without the explicit rule-based structure used to generate the structured task. We train a standard meta-learning agent, a recurrent network trained with model-free reinforcement learning, and compare it with human performance across the two task distributions. We find a double dissociation in which humans do better in the structured task distribution whereas agents do better in the null task distribution -- despite comparable statistical complexity. This work highlights that multiple strategies can achieve reasonable meta-test performance, and that careful construction of control task distributions is a valuable way to understand which strategies meta-learners acquire, and how they might differ from humans. | Accept (Poster) |
ICLR.cc/2023/Conference | Attention Enables Zero Approximation Error | Attention-based architectures become the core backbone of many state-of-the-art models for various tasks, including language translation and image classification. However, theoretical properties of attention-based models are seldom considered. In this work, we show that with suitable adaptations, the single-head self-attention transformer with a fixed number of transformer encoder blocks and free parameters is able to generate any desired polynomial of the input with no error. The number of transformer encoder blocks is the same as the degree of the target polynomial. Even more exciting, we find that these transformer encoder blocks in this model do not need to be trained. As a direct consequence, we show that the single-head self-attention transformer with increasing numbers of free parameters is universal. Also, we show that our proposed model can avoid the classical trade-off between approximation error and sample error in the mean squared error analysis for the regression task if the target function is a polynomial. We conduct various experiments and ablation studies to verify our theoretical results. | Reject |
ICLR.cc/2022/Conference | Convergence of Generalized Belief Propagation Algorithm on Graphs with Motifs | Belief propagation is a fundamental message-passing algorithm for numerous applications in machine learning. It is known that belief propagation algorithm is exact on tree graphs. However, belief propagation is run on loopy graphs in most applications. So, understanding the behavior of belief propagation on loopy graphs has been a major topic for researchers in different areas. In this paper, we study the convergence behavior of generalized belief propagation algorithm on graphs with motifs (triangles, loops, etc.) We show under a certain initialization, generalized belief propagation converges to the global optimum of the Bethe free energy for ferromagnetic Ising models on graphs with motifs. | Reject |
ICLR.cc/2022/Conference | Mask and Understand: Evaluating the Importance of Parameters | Influence functions are classic techniques from robust statistics based on first-order Taylor approximations that have been widely used in the machine learning community to estimate small perturbations of datasets accurately to the model. However, existing researches concentrate on the estimate the perturbations of the training or pre-training points. In this paper, we introduce the influence functions to predict the effects of removing features or parameters. It is worth emphasizing that our method can be applied to explore the influence of any combination of parameters disturbance on the model whether they belong to the same layer or whether are related. The validation and experiments also demonstrate that the influence functions for parameters can be used in many fields such as understanding model structure, model pruning, feature importance ranking, and any other strategies of masking parameters as you can imagine when you want to evaluate the importance of a group of parameters. | Withdrawn |
ICLR.cc/2023/Conference | Incremental Predictive Coding: A Parallel and Fully Automatic Learning Algorithm | Neuroscience-inspired models, such as predictive coding, have the potential to play an important role in the future of machine intelligence. However, they are not yet used in industrial applications due to some limitations, such as efficiency. In this work, we propose incremental predictive coding (iPC), a variation of the original model derived from the incremental expectation maximization algorithm, where every operation can be performed in parallel without external control. We show both theoretically and empirically that iPC is more efficient than the original algorithm by Rao and Ballard, with performances comparable to those of backpropagation in image classification tasks. This work impacts several areas, as it has general applications in computational neuroscience and machine learning, and specific applications in scenarios where automatization and parallelization are important, such as distributed computing and implementations of deep learning models on analog and neuromorphic chips. | Reject |
ICLR.cc/2019/Conference | Meta-Learning to Guide Segmentation | There are myriad kinds of segmentation, and ultimately the `"right" segmentation of a given scene is in the eye of the annotator. Standard approaches require large amounts of labeled data to learn just one particular kind of segmentation. As a first step towards relieving this annotation burden, we propose the problem of guided segmentation: given varying amounts of pixel-wise labels, segment unannotated pixels by propagating supervision locally (within an image) and non-locally (across images). We propose guided networks, which extract a latent task representation---guidance---from variable amounts and classes (categories, instances, etc.) of pixel supervision and optimize our architecture end-to-end for fast, accurate, and data-efficient segmentation by meta-learning. To span the few-shot and many-shot learning regimes, we examine guidance from as little as one pixel per concept to as much as 1000+ images, and compare to full gradient optimization at both extremes. To explore generalization, we analyze guidance as a bridge between different levels of supervision to segment classes as the union of instances. Our segmentor concentrates different amounts of supervision of different types of classes into an efficient latent representation, non-locally propagates this supervision across images, and can be updated quickly and cumulatively when given more supervision. | Reject |
ICLR.cc/2022/Conference | Omni-Dimensional Dynamic Convolution | Learning a single static convolutional kernel in each convolutional layer is the common training paradigm of modern Convolutional Neural Networks (CNNs). Instead, recent research in dynamic convolution shows that learning a linear combination of n convolutional kernels weighted with their input-dependent attentions can significantly improve the accuracy of light-weight CNNs, while maintaining efficient inference. However, we observe that existing works endow convolutional kernels with the dynamic property through one dimension (regarding the convolutional kernel number) of the kernel space, but the other three dimensions (regarding the spatial size, the input channel number and the output channel number for each convolutional kernel) are overlooked. Inspired by this, we present Omni-dimensional Dynamic Convolution (ODConv), a more generalized yet elegant dynamic convolution design, to advance this line of research. ODConv leverages a novel multi-dimensional attention mechanism with a parallel strategy to learn complementary attentions for convolutional kernels along all four dimensions of the kernel space at any convolutional layer. As a drop-in replacement of regular convolutions, ODConv can be plugged into many CNN architectures. Extensive experiments on the ImageNet and MS-COCO datasets show that ODConv brings solid accuracy boosts for various prevailing CNN backbones including both light-weight and large ones, e.g., 3.77%~5.71%|1.86%~3.72% absolute top-1 improvements to MobivleNetV2|ResNet family on the ImageNet dataset. Intriguingly, thanks to its improved feature learning ability, ODConv with even one single kernel can compete with or outperform existing dynamic convolution counterparts with multiple kernels, substantially reducing extra parameters. Furthermore, ODConv is also superior to other attention modules for modulating the output features or the convolutional weights. Code and models will be available at https://github.com/OSVAI/ODConv. | Accept (Spotlight) |
ICLR.cc/2021/Conference | Towards Robust Neural Networks via Close-loop Control | Despite their success in massive engineering applications, deep neural networks are vulnerable to various perturbations due to their black-box nature. Recent study has shown that a deep neural network can misclassify the data even if the input data is perturbed by an imperceptible amount. In this paper, we address the robustness issue of neural networks by a novel close-loop control method from the perspective of dynamic systems. Instead of modifying the parameters in a fixed neural network architecture, a close-loop control process is added to generate control signals adaptively for the perturbed or corrupted data. We connect the robustness of neural networks with optimal control using the geometrical information of underlying data to design the control objective. The detailed analysis shows how the embedding manifolds of state trajectory affect error estimation of the proposed method. Our approach can simultaneously maintain the performance on clean data and improve the robustness against many types of data perturbations. It can also further improve the performance of robustly trained neural networks against different perturbations. To the best of our knowledge, this is the first work that improves the robustness of neural networks with close-loop control. | Accept (Poster) |
ICLR.cc/2023/Conference | Fast Yet Effective Graph Unlearning through Influence Analysis | Recent evolving data privacy policies and regulations have led to increasing interest in the machine unlearning problem. In this paper, we consider Graph Neural Networks (GNNs) as the target model, and study the problem of edge unlearning in GNNs, i.e., learning a new GNN model as if a specified set of edges never existed in the original training graph. Despite its practical importance, the problem remains elusive due to the non-convexity nature of GNNs. Our main technical contribution is three-fold: 1) we cast the problem of edge unlearning as estimating the influence functions of the edges to be removed; 2) we design a computationally and memory efficient algorithm named EraEdge for edge influence estimation and unlearning; 3) under standard regularity conditions, we prove that the sequence of iterates produced by our algorithm converges to the desired model. A comprehensive set of experiments on three prominent GNN models and four benchmark graph datasets demonstrate that our algorithm achieves significant speed-up gains over retraining from scratch without sacrificing the model accuracy too much. Furthermore, our algorithm outperforms the existing GNN unlearning approach in terms of both training time and accuracy of the target GNN model. | Reject |
ICLR.cc/2023/Conference | Understanding and Mitigating Robust Overfitting through the Lens of Feature Dynamics | Adversarial Training (AT) has become arguably the state-of-the-art algorithm for extracting robust features. However, researchers recently notice that AT suffers from severe robust overfitting problems, particularly after the learning rate (LR) decay, while the existing static view of feature robustness fails to explain this phenomenon. In this paper, we propose a new dynamic feature robustness framework which takes the dynamic interplay between the model trainer and the attacker into consideration. By tracing temporal and dataset-specific feature robustness, we develop a new understanding of robust overfitting from the dynamics of non-robust features, and empirically verify it on real-world datasets. Built upon this understanding, we explore three techniques to restore the balance between the model trainer and the attacker, and show that they could effectively alleviate robust overfitting and attain state-of-the-art robustness on benchmark datasets. Notably, different from previous studies, our interpretation highlights the necessity of considering the min-max nature of AT for robust overfitting.
| Reject |
ICLR.cc/2023/Conference | Characterizing intrinsic compositionality in transformers with Tree Projections | When trained on language data, do transformers learn some arbitrary computation that utilizes the full capacity of the architecture or do they learn a simpler, tree-like computation, hypothesized to underlie compositional meaning systems like human languages? There is an apparent tension between compositional accounts of human language understanding, which are based on a restricted bottom-up computational process, and the enormous success of neural models like transformers, which can route information arbitrarily between different parts of their input. One possibility is that these models, while extremely flexible in principle, in practice learn to interpret language hierarchically, ultimately building sentence representations close to those predictable by a bottom-up, tree-structured model. To evaluate this possibility, we describe an unsupervised and parameter-free method to \emph{functionally project} the behavior of any transformer into the space of tree-structured networks. Given an input sentence, we produce a binary tree that approximates the transformer's representation-building process and a score that captures how ``tree-like'' the transformer's behavior is on the input. While calculation of this score does not require training any additional models, it provably upper-bounds the fit between a transformer and any tree-structured approximation. Using this method, we show that transformers for three different tasks become more tree-like over the course of training, in some cases unsupervisedly recovering the same trees as supervised parsers. These trees, in turn, are predictive of model behavior, with more tree-like models generalizing better on tests of compositional generalization. | Accept: poster |
ICLR.cc/2023/Conference | What Makes Convolutional Models Great on Long Sequence Modeling? | Convolutional models have been widely used in multiple domains. However, most existing models only use local convolution, making the model unable to handle long-range dependencies efficiently. Attention overcomes this problem by aggregating global information based on the pair-wise attention score but also makes the computational complexity quadratic to the sequence length. Recently, Gu et al. proposed a model called S4 inspired by the state space model. S4 can be efficiently implemented as a global convolutional model whose kernel size equals the input sequence length. With Fast Fourier Transform, S4 can model much longer sequences than Transformers and achieve significant gains over SoTA on several long-range tasks. Despite its empirical success, S4 is involved. It requires sophisticated parameterization and initialization schemes that combine the wisdom from several prior works. As a result, S4 is less intuitive and hard to use for researchers with limited prior knowledge. Here we aim to demystify S4 and extract basic principles that contribute to the success of S4 as a global convolutional model. We focus on the structure of the convolution kernel and identify two critical but intuitive principles enjoyed by S4 that are sufficient to make up an effective global convolutional model: 1) The parameterization of the convolutional kernel needs to be efficient in the sense that the number of parameters should scale sub-linearly with sequence length. 2) The kernel needs to satisfy a decaying structure that the weights for convolving with closer neighbors are larger than the more distant ones. Based on the two principles, we propose a simple yet effective convolutional model called Structured Global Convolution (SGConv). SGConv exhibits strong empirical performance over several tasks: 1) With faster speed, SGConv surpasses the previous SoTA on Long Range Arena and Speech Command datasets. 2) When plugging SGConv into standard language and vision models, it shows the potential to improve both efficiency and performance. | Accept: poster |
ICLR.cc/2021/Conference | Adversarial Training using Contrastive Divergence | To protect the security of machine learning models against adversarial examples, adversarial training becomes the most popular and powerful strategy against various adversarial attacks by injecting adversarial examples into training data. However, it is time-consuming and requires high computation complexity to generate suitable adversarial examples for ensuring the robustness of models, which impedes the spread and application of adversarial training. In this work, we reformulate adversarial training as a combination of stationary distribution exploring, sampling, and training. Each updating of parameters of DNN is based on several transitions from the data samples as the initial states in a Hamiltonian system. Inspired by our new paradigm, we design a new generative method for adversarial training by using Contrastive Divergence (ATCD), which approaches the equilibrium distribution of adversarial examples with only few iterations by building from small modifications of the standard Contrastive Divergence (CD). Our adversarial training algorithm achieves much higher robustness than any other state-of-the-art adversarial training acceleration method on the ImageNet, CIFAR-10, and MNIST datasets and reaches a balance between performance and efficiency. | Reject |
ICLR.cc/2022/Conference | VC dimension of partially quantized neural networks in the overparametrized regime | Vapnik-Chervonenkis (VC) theory has so far been unable to explain the small generalization error of overparametrized neural networks. Indeed, existing applications of VC theory to large networks obtain upper bounds on VC dimension that are proportional to the number of weights, and for a large class of networks, these upper bound are known to be tight. In this work, we focus on a class of partially quantized networks that we refer to as hyperplane arrangement neural networks (HANNs). Using a sample compression analysis, we show that HANNs can have VC dimension significantly smaller than the number of weights, while being highly expressive. In particular, empirical risk minimization over HANNs in the overparametrized regime achieves the minimax rate for classification with Lipschitz posterior class probability. We further demonstrate the expressivity of HANNs empirically. On a panel of 121 UCI datasets, overparametrized HANNs are able to match the performance of state-of-the-art full-precision models. | Accept (Poster) |
ICLR.cc/2021/Conference | Measuring GAN Training in Real Time | Generative Adversarial Networks (GAN) are popular generative models of images. Although researchers proposed variants of GAN for different applications, evaluating and comparing GANs is still challenging as GANs may have many failure cases such as low visual quality and model collapse. To alleviate this issue, we propose a novel framework to evaluate the training stability (S), visual quality (Q), and mode diversity (D) of GAN simultaneously. SQD requires only a moderate number of samples, allowing real-time monitoring of the training dynamics of GAN. We showcase the utility of the SQD framework on prevalent GANs and discovered that the gradient penalty (Gulrajani et al., 2017) regularization significantly improves the performance of GAN. We also compare the gradient penalty regularization with other regularization methods and reveal that enforcing the 1-Lipschitz condition of the discriminator network stabilizes GAN training. | Withdrawn |
ICLR.cc/2023/Conference | Show and Write: Entity-aware Article Generation with Image Information | Prior work for article generation has primarily focused on generating articles using a human-written prompt to provide topical context and metadata about the article. However, for many applications, such as generating news stories, these articles are also often paired with images and their captions or alt-text, which in turn are based on real-world events and may reference many different named entities that are difficult to be correctly recognized and predicted by language models. To address this shortcoming, this paper introduces an ENtity-aware article Generation method with Image iNformation, ENGIN, to incorporate an article's image information into language models. ENGIN represents articles that can be conditioned on metadata used by prior work and information such as captions and named entities extracted from images. Our key contribution is a novel Entity-aware mechanism to help our model recognize and predict the entity names in articles. We perform experiments on three public datasets, GoodNews, VisualNews, and WikiText. Quantitative results show that our approach improves generated article perplexity by 4-5 points over the base models. Qualitative results demonstrate the text generated by ENGIN is more consistent with embedded article images. We also perform article quality annotation experiments on the generated articles to validate that our model produces higher-quality articles. Finally, we investigate the effect ENGIN has on methods that automatically detect machine-generated articles. | Reject |
ICLR.cc/2021/Conference | Max-Affine Spline Insights Into Deep Network Pruning | In this paper, we study the importance of pruning in Deep Networks (DNs) and motivate it based on the current absence of data aware weight initialization. Current DN initializations, focusing primarily at maintaining first order statistics of the feature maps through depth, force practitioners to overparametrize a model in order to reach high performances. This overparametrization can then be pruned a posteriori, leading to a phenomenon known as "winning tickets". However, the pruning literature still relies on empirically investigations, lacking a theoretical understanding of (1) how pruning affects the decision boundary, (2) how to interpret pruning, (3) how to design principled pruning techniques, and (4) how to theoretically study pruning. To tackle those questions, we propose to employ recent advances in theoretical analysis of Continuous Piecewise Affine (CPA) DNs. From this viewpoint, we can study the DNs' input space partitioning and detect the early-bird (EB) phenomenon, guide practitioners by identifying when to stop the first training step, provide interpretability into current pruning techniques, and develop a principled pruning criteria towards efficient DN training. Finally, we conduct extensive experiments to shown the effectiveness of the proposed spline pruning criteria in terms of both layerwise and global pruning over state-of-the-art pruning methods.
All the codes will be released publicly upon acceptance. | Withdrawn |
ICLR.cc/2022/Conference | T-WaveNet: A Tree-Structured Wavelet Neural Network for Time Series Signal Analysis | Time series signal analysis plays an essential role in many applications, e.g., activity recognition and healthcare monitoring.
Recently, features extracted with deep neural networks (DNNs) have shown to be more effective than conventional hand-crafted ones.
However, most existing solutions rely solely on the network to extract information carried in the raw signal, regardless of its inherent physical and statistical properties, leading to sub-optimal performance particularly under a limited amount of training data.
In this work, we propose a novel tree-structured wavelet neural network for time series signal analysis, namely \emph{T-WaveNet}, taking advantage of an inherent property of various types of signals, known as the \emph{dominant frequency range}. Specifically, with \emph{T-WaveNet}, we first conduct frequency spectrum energy analysis of the signals to get a set of dominant frequency subbands. Then, we construct a tree-structured network that iteratively decomposes the input signal into various frequency subbands with similar energies. Each node on the tree is built with an invertible neural network (INN) based wavelet transform unit. Such a disentangled representation learning method facilitates a more effective extraction of the discriminative features, as demonstrated with the comprehensive experiments on various real-life time series classification datasets. | Accept (Poster) |
ICLR.cc/2022/Conference | Churn Reduction via Distillation | In real-world systems, models are frequently updated as more data becomes available, and in addition to achieving high accuracy, the goal is to also maintain a low difference in predictions compared to the base model (i.e. predictive churn). If model retraining results in vastly different behavior, then it could cause negative effects in downstream systems, especially if this churn can be avoided with limited impact on model accuracy. In this paper, we show an equivalence between training with distillation using the base model as the teacher and training with an explicit constraint on the predictive churn. We then show that distillation performs strongly for low churn training against a number of recent baselines on a wide range of datasets and model architectures, including fully-connected networks, convolutional networks, and transformers. | Accept (Spotlight) |
ICLR.cc/2023/Conference | UNDERSTANDING THE ROLE OF POSITIONAL ENCODINGS IN SENTENCE REPRESENTATIONS | Positional encodings are used to inject word-order information into transformer-based language models. While they can significantly enhance the quality of sentence representations, their specific contribution to language models are not fully understood, especially given recent findings that building natural-language understanding from language models with positional encodings is insensitive to word order. In this work, we investigate the role of positional encodings systematically. (1) We uncover the core function of existing positional encodings is to symmetrically combine local units by identifying two common properties, Locality, and Symmetry. (2) We reveal that positional and contextual encodings play a distinct role in understanding sentences. (3) Based on these findings, we propose a simplified new method to inject positional information into such models. Empirical studies demonstrate that this method can improve the performance of the BERT-based model on 10 downstream tasks. We hope these new probing results and findings can shed light on how to design and inject positional encodings into language models.
| Reject |
ICLR.cc/2023/Conference | Dynamic Loss for Learning with Label Noise | Label noise is verified seriously harmful to deep neural networks (DNNs). A simple and scalable strategy to handle this problem is to design robust loss functions, which improve generalization in the presence of label noise by reconciling fitting ability with robustness. However, the widely-used static trade-off between the two contradicts the dynamics of DNNs learning with label noise, leading to an inferior performance. Therefore, in this paper, we propose a dynamic loss function to solve this problem. Specifically, DNNs tend to first learn generalized patterns, then gradually overfit label noise. In light of this, we make fitting ability stronger initially, then gradually increase the weight of robustness. Moreover, we let DNNs put more emphasis on easy examples than hard ones at the later stage since the former are correctly labeled with a higher probability, further reducing the negative impact of label noise. Extensive experimental results on various benchmark datasets demonstrate the state-of-the-art performance of our method. We will open-source our code very soon. | Withdrawn |
ICLR.cc/2022/Conference | A fast and accurate splitting method for optimal transport: analysis and implementation | We develop a fast and reliable method for solving large-scale optimal transport (OT) problems at an unprecedented combination of speed and accuracy. Built on the celebrated Douglas-Rachford splitting technique, our method tackles the original OT problem directly instead of solving an approximate regularized problem, as many state-of-the-art techniques do. This allows us to provide sparse transport plans and avoid numerical issues of methods that use entropic regularization. The algorithm has the same cost per iteration as the popular Sinkhorn method, and each iteration can be executed efficiently, in parallel. The proposed method enjoys an iteration complexity $O(1/\epsilon)$ compared to the best-known $O(1/\epsilon^2)$ of the Sinkhorn method. In addition, we establish a linear convergence rate for our formulation of the OT problem. We detail an efficient GPU implementation of the proposed method that maintains a primal-dual stopping criterion at no extra cost. Substantial experiments demonstrate the effectiveness of our method, both in terms of computation times and robustness. | Accept (Poster) |
ICLR.cc/2021/Conference | Revealing the Structure of Deep Neural Networks via Convex Duality | We study regularized deep neural networks (DNNs) and introduce a convex analytic framework to characterize the structure of the hidden layers. We show that a set of optimal hidden layer weights for a norm regularized DNN training problem can be explicitly found as the extreme points of a convex set. For the special case of deep linear networks with $K$ outputs, we prove that each optimal weight matrix is rank-$K$ and aligns with the previous layers via duality. More importantly, we apply the same characterization to deep ReLU networks with whitened data and prove the same weight alignment holds. As a corollary, we prove that norm regularized deep ReLU networks yield spline interpolation for one-dimensional datasets which was previously known only for two-layer networks. Furthermore, we provide closed-form solutions for the optimal layer weights when data is rank-one or whitened. We then verify our theory via numerical experiments. | Reject |
ICLR.cc/2023/Conference | Vectorial Graph Convolutional Networks | Graph Convolutional Networks (GCN) have drawn considerable attention recently due to their outstanding performance in processing graph-structured data. However, GCNs still limited to the undirected graph because they theoretically require a symmetric matrix as the basis for the Laplacian transform. This causes the isotropic problem of the operator and reduced sensitivity in response to different information. In order to solve the problem, we generalize the spectral convolution operator to directed graphs by field extension, which improves the edge representations from scalars to vectors. Therefore, it brings in the concept of direction. That is to say, and even homogeneous information can become distinguishable by its differences in directions.In this paper, we propose the Vectorial Graph Convolutional Network(VecGCN) and the experimental evidence showing the advantages of a variety of directed graph node classification and link prediction tasks. | Reject |
ICLR.cc/2023/Conference | Toxicity in Multilingual Machine Translation at Scale | Machine Translation systems can produce different types of errors, some of which get characterized as critical or catastrophic due to the specific negative impact they can have on users. Automatic or human evaluation metrics do not necessarily differentiate between such critical errors and more innocuous ones. In this paper we focus on one type of critical error: added toxicity. We evaluate and analyze added toxicity when translating a large evaluation dataset (HOLISTICBIAS, over 472k sentences, covering 13 demographic axes) from English into 164 languages. The toxicity automatic evaluation shows that added toxicity across languages varies from 0% to 5%. The output languages with the most added toxicity tend to be low-resource ones, and the demographic axes with the most added toxicity include sexual orientation, gender and sex, and ability. We also perform human evaluation on a subset of 8 directions, confirming the prevalence of true
added toxicity.
We use a measurement of the amount of source contribution to the translation, where a low source contribution implies hallucination, to interpret what causes toxicity. We observe that the source contribution is somewhat correlated with toxicity but that 45.6% of added toxic words have a high source contribution, suggesting that much of the added toxicity may be due to mistranslations. Combining the signal of source contribution level with a measurement of translation robustness allows us to flag 22.3% of added toxicity, suggesting that added toxicity may be related to both hallucination and the stability of translations in different contexts. Given these findings, our recommendations to reduce added toxicity are to curate training data to avoid mistranslations, mitigate hallucination and check unstable translations. | Reject |
ICLR.cc/2022/Conference | Revisiting and Advancing Fast Adversarial Training Through the lens of Bi-Level Optimization | Adversarial training (AT) has become a widely recognized defense mechanism to improve the robustness of deep neural networks against adversarial attacks. It is originated from solving a min-max optimization problem, where the minimizer (i.e., defender) seeks a robust model to minimize the worst-case training loss at the presence of adversarial examples crafted by the maximizer (i.e., attacker). However,the min-max nature makes AT computationally intensive and thus difficult to scale. Thus, the problem of FAST-AT arises. Nearly all the recent progress is achieved based on the following simplification: The iterative attack generation method used in the maximization step of AT is replaced by the simplest one-shot gradient sign-based PGD method. Nevertheless, FAST-AT is far from satisfactory, and it lacks theoretically-grounded design. For example, a FAST-AT method may suffer from robustness catastrophic overfitting when training with strong adversaries.
In this paper, we foster a technological breakthrough for designing FAST-AT through the lens of bi-level optimization (BLO) instead of min-max optimization. First, we theoretically show that the most commonly-used algorithmic specification of FAST-AT is equivalent to the linearized BLO along the direction given by the sign of input gradient. Second, with the aid of BLO, we develop a new systematic and effective fast bi-level AT framework, termed FAST-BAT, whose algorithm is rigorously derived by leveraging the theory of implicit gradient. In contrast to FAST-AT, FAST-BAT has the least restriction to placing the tradeoff between computation efficiency and adversarial robustness. For example, it is capable of defending sign-based projected gradient descent (PGD) attacks without calling any gradient sign method and explicit robust regularization during training. Furthermore, we empirically show that our method outperforms state-of-the-art FAST-AT baselines. In particular, FAST-BAT can achieve superior model robustness without inducing robustness catastrophic overfitting and losing standard accuracy. | Reject |
ICLR.cc/2023/Conference | SIMPLE: A Gradient Estimator for k-Subset Sampling | $k$-subset sampling is ubiquitous in machine learning, enabling regularization and interpretability through sparsity. The challenge lies in rendering $k$-subset sampling amenable to end-to-end learning. This has typically involved relaxing the reparameterized samples to allow for backpropagation, but introduces both bias and variance. In this work, we fall back to discrete $k$-subset sampling on the forward pass. This is coupled with using the gradient with respect to the exact marginals, computed efficiently, as a proxy for the true gradient. We show that our gradient estimator exhibits lower bias and variance compared to state-of-the-art estimators. Empirical results show improved performance on learning to explain and sparse models benchmarks. We provide an algorithm for computing the exact ELBO for the $k$-subset distribution, obtaining significantly lower loss compared to state-of-the-art discrete sparse VAEs. All of our algorithms are exact and efficient. | Accept: poster |
ICLR.cc/2023/Conference | Triangle Inequality for Inverse Optimal Control | Inverse optimal control (IOC) is a problem of estimating a cost function based on the behaviors of an expert that behaves optimally with respect to the cost function. Although the Hamilton-Jacobi-Bellman (HJB) equation for the value function that evaluates the temporal integral of the cost function provides a necessary condition for the optimality of expert behaviors, the use of the HJB equation alone is insufficient for solving the IOC problem. In this study, we propose a triangle inequality which is useful for estimating the better representation of the value function, along with a new IOC method incorporating the triangle inequality. Through several IOC problems and imitation learning problems of time-dependent control behaviors, we show that our IOC method performs substantially better than an existing IOC method. Showing our IOC method is also applicable to an imitation of expert control of a 2-link manipulator, we demonstrate applicability of our method to real-world problems. | Withdrawn |
ICLR.cc/2021/Conference | Improving Learning to Branch via Reinforcement Learning | Branch-and-Bound~(B\&B) is a general and widely used algorithm paradigm for solving Mixed Integer Programming~(MIP).
Recently there is a surge of interest in designing learning-based branching policies as a fast approximation of strong branching, a human-designed heuristic. In this work, we argue strong branching is not a good expert to imitate for its poor decision quality when turning off its side effects in solving linear programming. To obtain more effective and non-myopic policies than a local heuristic, we formulate the branching process in MIP as reinforcement learning~(RL) and design a policy characterization for the B\&B process to improve our agent by novelty search evolutionary strategy. Across a range of NP-hard problems, our trained RL agent significantly outperforms expert-designed branching rules and the state-of-the-art learning-based branching methods in terms of both speed and effectiveness. Our results suggest that with carefully designed policy networks and learning algorithms, reinforcement learning has the potential to advance algorithms for solving MIPs. | Reject |
ICLR.cc/2020/Conference | DEEP GRAPH SPECTRAL EVOLUTION NETWORKS FOR GRAPH TOPOLOGICAL TRANSFORMATION | Characterizing the underlying mechanism of graph topological evolution from a source graph to a target graph has attracted fast increasing attention in the deep graph learning domain. However, there lacks expressive and efficient that can handle global and local evolution patterns between source and target graphs. On the other hand, graph topological evolution has been investigated in the graph signal processing domain historically, but it involves intensive labors to manually determine suitable prescribed spectral models and prohibitive difficulty to fit their potential combinations and compositions. To address these challenges, this paper proposes the deep Graph Spectral Evolution Network (GSEN) for modeling the graph topology evolution problem by the composition of newly-developed generalized graph kernels. GSEN can effectively fit a wide range of existing graph kernels and their combinations and compositions with the theoretical guarantee and experimental verification. GSEN has outstanding efficiency in terms of time complexity ($O(n)$) and parameter complexity ($O(1)$), where $n$ is the number of nodes of the graph. Extensive experiments on multiple synthetic and real-world datasets have demonstrated outstanding performance. | Reject |
ICLR.cc/2023/Conference | TextShield: Beyond Successfully Detecting Adversarial Sentences in text classification | Adversarial attack serves as a major challenge for neural network models in NLP, which precludes the model's deployment in safety-critical applications. A recent line of work, detection-based defense, aims to distinguish adversarial sentences from benign ones. However, {the core limitation of previous detection methods is being incapable of giving correct predictions on adversarial sentences unlike defense methods from other paradigms.} To solve this issue, this paper proposes TextShield: (1) we discover a link between text attack and saliency information, and then we propose a saliency-based detector, which can effectively detect whether an input sentence is adversarial or not. (2) We design a saliency-based corrector, which converts the detected adversary sentences to benign ones. By combining the saliency-based detector and corrector, TextShield extends the detection-only paradigm to a detection-correction paradigm, thus filling the gap in the existing detection-based defense. Comprehensive experiments show that (a) TextShield consistently achieves higher or comparable performance than state-of-the-art defense methods across various attacks on different benchmarks. (b) our saliency-based detector outperforms existing detectors for detecting adversarial sentences. | Accept: poster |
ICLR.cc/2023/Conference | Distributionally Robust Recourse Action | A recourse action aims to explain a particular algorithmic decision by showing one specific way in which the instance could be modified to receive an alternate outcome. Existing recourse generation methods often assume that the machine learning model does not change over time. However, this assumption does not always hold in practice because of data distribution shifts, and in this case, the recourse action may become invalid. To redress this shortcoming, we propose the Distributionally Robust Recourse Action (DiRRAc) framework, which generates a recourse action that has high probability of being valid under a mixture of model shifts. We first formulate the robustified recourse setup as a min-max optimization problem, where the max problem is specified by Gelbrich distance over an ambiguity set around the distribution of model parameters. Then we suggest a projected gradient descent algorithm to find a robust recourse according to the min-max objective. We also show that our DiRRAc framework can be extended to hedge against the misspecification of the mixture weights. Numerical experiments with both synthetic and three real-world datasets demonstrate the benefits of our proposed framework over the state-of-the-art recourse methods, which generate robust recourses.
| Accept: poster |
ICLR.cc/2023/Conference | Integrating Episodic and Global Novelty Bonuses for Efficient Exploration | Exploration in environments which differ across episodes has received increasing attention in recent years. Current methods use some combination of global novelty bonuses, computed using the agent's entire training experience, and episodic novelty bonuses, computed using only experience from the current episode. However, the use of these two types of bonuses has been ad-hoc and poorly understood. In this work, we first shed light on the behavior these two kinds of bonuses on hard exploration tasks through easily interpretable examples. We find that the two types of bonuses succeed in different settings, with episodic bonuses being most effective when there is little shared structure between environments and global bonuses being effective when more structure is shared. We also find that combining the two bonuses leads to more robust behavior across both of these settings. Motivated by these findings, we then investigate different algorithmic choices for defining and combining function approximation-based global and episodic bonuses. This results in a new algorithm which sets a new state of the art across 18 tasks from the MiniHack suite used in prior work. Our code is public at \url{web-link}. | Reject |
ICLR.cc/2020/Conference | Generating Robust Audio Adversarial Examples using Iterative Proportional Clipping | Audio adversarial examples, imperceptible to humans, have been constructed to attack automatic speech recognition (ASR) systems. However, the adversarial examples generated by existing approaches usually involve notable noise, especially during the periods of silence and pauses, which may lead to the detection of such attacks. This paper proposes a new approach to generate adversarial audios using Iterative Proportional Clipping (IPC), which exploits temporal dependency in original audios to significantly limit human-perceptible noise. Specifically, in every iteration of optimization, we use a backpropagation model to learn the raw perturbation on the original audio to construct our clipping. We then impose a constraint on the perturbation at the positions with lower sound intensity across the time domain to eliminate the perceptible noise during the silent periods or pauses. IPC preserves the linear proportionality between the original audio and the perturbed one to maintain the temporal dependency. We show that the proposed approach can successfully attack the latest state-of-the-art ASR model Wav2letter+, and only requires a few minutes to generate an audio adversarial example. Experimental results also demonstrate that our approach succeeds in preserving temporal dependency and can bypass temporal dependency based defense mechanisms. | Reject |
ICLR.cc/2021/Conference | Learning to Solve Nonlinear Partial Differential Equation Systems To Accelerate MOSFET Simulation | Semiconductor device simulation uses numerical analysis, where a set of coupled nonlinear partial differential equations is solved with the iterative Newton-Raphson method. Since an appropriate initial guess to start the Newton-Raphson method is not available, a solution of practical importance with desired boundary conditions cannot be trivially achieved. Instead, several solutions with intermediate boundary conditions should be calculated to address the nonlinearity and introducing intermediate boundary conditions significantly increases the computation time. In order to accelerate the semiconductor device simulation, we propose to use a neural network to learn an approximate solution for desired boundary conditions. With an initial solution sufficiently close to the final one by a trained neural network, computational cost to calculate several unnecessary solutions is significantly reduced. Specifically, a convolutional neural network for MOSFET (Metal-Oxide-Semiconductor Field-Effect Transistor), the most widely used semiconductor device, are trained in a supervised manner to compute the initial solution. Particularly, we propose to consider device grids with varying size and spacing and derive a compact expression of the solution based upon the electrostatic potential. We empirically show that the proposed method accelerates the simulation by more than 12 times. Results from the local linear regression and a fully-connected network are compared and extension to a complex two-dimensional domain is sketched. | Reject |
ICLR.cc/2020/Conference | Coresets for Accelerating Incremental Gradient Methods | Many machine learning problems reduce to the problem of minimizing an expected risk. Incremental gradient (IG) methods, such as stochastic gradient descent and its variants, have been successfully used to train the largest of machine learning models. IG methods, however, are in general slow to converge and sensitive to stepsize choices. Therefore, much work has focused on speeding them up by reducing the variance of the estimated gradient or choosing better stepsizes. An alternative strategy would be to select a carefully chosen subset of training data, train only on that subset, and hence speed up optimization. However, it remains an open question how to achieve this, both theoretically as well as practically, while not compromising on the quality of the final model. Here we develop CRAIG, a method for selecting a weighted subset (or coreset) of training data in order to speed up IG methods. We prove that by greedily selecting a subset S of training data that minimizes the upper-bound on the estimation error of the full gradient, running IG on this subset will converge to the (near)optimal solution in the same number of epochs as running IG on the full data. But because at each epoch the gradients are computed only on the subset S, we obtain a speedup that is inversely proportional to the size of S. Our subset selection algorithm is fully general and can be applied to most IG methods. We further demonstrate practical effectiveness of our algorithm, CRAIG, through an extensive set of experiments on several applications, including logistic regression and deep neural networks. Experiments show that CRAIG, while achieving practically the same loss, speeds up IG methods by up to 10x for convex and 3x for non-convex (deep learning) problems. | Reject |
ICLR.cc/2020/Conference | Exploiting Excessive Invariance caused by Norm-Bounded Adversarial Robustness | Adversarial examples are malicious inputs crafted to cause a model to misclassify them. In their most common instantiation, "perturbation-based" adversarial examples introduce changes to the input that leave its true label unchanged, yet result in a different model prediction. Conversely, "invariance-based" adversarial examples insert changes to the input that leave the model's prediction unaffected despite the underlying input's label having changed. So far, the relationship between these two notions of adversarial examples has not been studied, we close this gap.
We demonstrate that solely achieving perturbation-based robustness is insufficient for complete adversarial robustness. Worse, we find that classifiers trained to be Lp-norm robust are more vulnerable to invariance-based adversarial examples than their undefended counterparts. We construct theoretical arguments and analytical examples to justify why this is the case. We then illustrate empirically that the consequences of excessive perturbation-robustness can be exploited to craft new attacks. Finally, we show how to attack a provably robust defense --- certified on the MNIST test set to have at least 87% accuracy (with respect to the original test labels) under perturbations of Linfinity-norm below epsilon=0.4 --- and reduce its accuracy (under this threat model with respect to an ensemble of human labelers) to 60% with an automated attack, or just 12% with human-crafted adversarial examples. | Reject |
ICLR.cc/2021/Conference | Uncertainty Estimation and Calibration with Finite-State Probabilistic RNNs | Uncertainty quantification is crucial for building reliable and trustable machine learning systems. We propose to estimate uncertainty in recurrent neural networks (RNNs) via stochastic discrete state transitions over recurrent timesteps. The uncertainty of the model can be quantified by running a prediction several times, each time sampling from the recurrent state transition distribution, leading to potentially different results if the model is uncertain. Alongside uncertainty quantification, our proposed method offers several advantages in different settings. The proposed method can (1) learn deterministic and probabilistic automata from data, (2) learn well-calibrated models on real-world classification tasks, (3) improve the performance of out-of-distribution detection, and (4) control the exploration-exploitation trade-off in reinforcement learning. An implementation is available. | Accept (Poster) |
ICLR.cc/2022/Conference | D$^2$ETR: Decoder-Only DETR with Computationally Efficient Cross-Scale Attention | DETR is the first fully end-to-end detector that predicts a final set of predictions without post-processing. However, it suffers from problems such as low performance and slow convergence. A series of works aim to tackle these issues in different ways, but the computational cost is yet expensive due to the sophisticated encoder-decoder architecture. To alleviate this issue, we propose a decoder-only detector called D$^2$ETR. In the absence of encoder, the decoder directly attends to the fine-fused feature maps generated by the Transformer backbone with a novel computationally efficient cross-scale attention module. D$^2$ETR demonstrates low computational complexity and high detection accuracy in evaluations on the COCO benchmark, outperforming DETR and its variants. | Withdrawn |
ICLR.cc/2022/Conference | Fooling Adversarial Training with Induction Noise | Adversarial training is widely believed to be a reliable approach to improve model robustness against adversarial attack. However, in this paper, we show that when trained on one type of poisoned data, adversarial training can also be fooled to have catastrophic behavior, e.g., $<1\%$ robust test accuracy with $>90\%$ robust training accuracy on CIFAR-10 dataset. Previously, there are other types of noise poisoned in the training data that have successfully fooled standard training ($15.8\%$ standard test accuracy with $99.9\%$ standard training accuracy on CIFAR-10 dataset), but their poisonings can be easily removed when adopting adversarial training. Therefore, we aim to design a new type of inducing noise, named ADVIN, which is an irremovable poisoning of training data. ADVIN can not only degrade the robustness of adversarial training by a large margin, for example, from $51.7\%$ to $0.57\%$ on CIFAR-10 dataset, but also be effective for fooling standard training ($13.1\%$ standard test accuracy with $100\%$ standard training accuracy). Additionally, ADVIN can be applied to preventing personal data (like selfies) from being exploited without authorization under whether standard or adversarial training. | Reject |
ICLR.cc/2023/Conference | gGN: learning to represent nodes in directed graphs as low-rank Gaussian distributions | Unsupervised learning of node representations from knowledge graphs is critical for numerous downstream tasks, ranging from large-scale graph analysis to measuring semantic similarity between nodes. This study presents gGN as a novel representation that defines graph nodes as Gaussian distributions. Unlike existing representations that approximate such distributions using diagonal covariance matrices, our proposal approximates them using low-rank perturbations. We demonstrate that this low-rank approximation is more expressive and better suited to represent complex asymmetric relations between nodes. In addition, we provide a computationally affordable algorithm for learning the low-rank representations in an unsupervised fashion. This learning algorithm uses a novel loss function based on the reverse Kullback-Leibler divergence and two ranking metrics whose joint minimization results in node representations that preserve not only node depths but also local and global asymmetric relationships between nodes. We assessed the representation power of the low-rank approximation with an in-depth systematic empirical study. The results show that our proposal was significantly better than the diagonal approximation for preserving graph structures. Moreover, gGN also outperformed 17 methods on the downstream task of measuring semantic similarity between graph nodes. | Withdrawn |
ICLR.cc/2023/Conference | Brainformers: Trading Simplicity for Efficiency | Transformers are central to recent successes in natural language processing and computer vision. Transformers have a mostly uniform backbone where layers alternate between feed-forward and self-attention in order to build a deep network. Here we investigate this design choice and find that more complex blocks that have different permutations of feed-forward layers, self-attention layers, and gated layers (for routing through sparse structures) can be more efficient. Using this insight, we develop a complex block, named Brainformer, that consists of a diverse sets of layers such as sparsely gated feed-forward layers with different gating mechanisms, dense feed-forward layers, attention layers, and various forms of layer normalization and activation functions. Brainformer consistently outperforms the state-of-the-art dense and sparse Transformers, in terms of both quality and efficiency. A Brainformer model with 8 billion activated parameters per token demonstrates 2$\times$ faster training convergence and 5x faster step time compared to its GLaM counterpart. In downstream task evaluation, Brainformer also demonstrates a 3% higher SuperGLUE score with fine-tuning compared to GLaM with a similar number of activated parameters. Finally, Brainformer largely outperforms a Primer dense model with similar computation per token on oneshot evaluation for three important generative tasks. | Withdrawn |
ICLR.cc/2018/Conference | Graph2Seq: Scalable Learning Dynamics for Graphs | Neural networks are increasingly used as a general purpose approach to learning algorithms over graph structured data. However, techniques for representing graphs as real-valued vectors are still in their infancy. Recent works have proposed several approaches (e.g., graph convolutional networks), but as we show in this paper, these methods have difficulty generalizing to large graphs. In this paper we propose Graph2Seq, an embedding framework that represents graphs as an infinite time-series. By not limiting the representation to a fixed dimension, Graph2Seq naturally scales to graphs of arbitrary size. Moreover, through analysis of a formal computational model we show that an unbounded sequence is necessary for scalability. Graph2Seq is also reversible, allowing full recovery of the graph structure from the sequence. Experimental evaluations of Graph2Seq on a variety of combinatorial optimization problems show strong generalization and strict improvement over state of the art. | Reject |
ICLR.cc/2023/Conference | Transformer Meets Boundary Value Inverse Problems | A Transformer-based deep direct sampling method is proposed for electrical impedance tomography, a well-known severely ill-posed nonlinear boundary value inverse problem. A real-time reconstruction is achieved by evaluating the learned inverse operator between carefully designed data and the reconstructed images. An effort is made to give a specific example to a fundamental question: whether and how one can benefit from the theoretical structure of a mathematical problem to develop task-oriented and structure-conforming deep neural networks? Specifically, inspired by direct sampling methods for inverse problems, the 1D boundary data in different frequencies are preprocessed by a partial differential equation-based feature map to yield 2D harmonic extensions as different input channels. Then, by introducing learnable non-local kernels, the direct sampling is recast to a modified attention mechanism. The new method achieves superior accuracy over its predecessors and contemporary operator learners and shows robustness to noises in benchmarks.
This research shall strengthen the insights that, despite being invented for natural language processing tasks, the attention mechanism offers great flexibility to be modified in conformity with the a priori mathematical knowledge, which ultimately leads to the design of more physics-compatible neural architectures. | Accept: poster |
ICLR.cc/2020/Conference | Regularization Matters in Policy Optimization | Deep Reinforcement Learning (Deep RL) has been receiving increasingly more attention thanks to its encouraging performance on a variety of control tasks. Yet, conventional regularization techniques in training neural networks (e.g., $L_2$ regularization, dropout) have been largely ignored in RL methods, possibly because agents are typically trained and evaluated in the same environment. In this work, we present the first comprehensive study of regularization techniques with multiple policy optimization algorithms on continuous control tasks. Interestingly, we find conventional regularization techniques on the policy networks can often bring large improvement on the task performance, and the improvement is typically more significant when the task is more difficult. We also compare with the widely used entropy regularization and find $L_2$ regularization is generally better. Our findings are further confirmed to be robust against the choice of training hyperparameters. We also study the effects of regularizing different components and find that only regularizing the policy network is typically enough. We hope our study provides guidance for future practices in regularizing policy optimization algorithms. | Reject |
ICLR.cc/2023/Conference | Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training | Running out of GPU memory has become a main bottleneck for large-scale DNN training. How to reduce the memory footprint during training has received intensive research attention. We find that previous gradient accumulation reduces activation memory but fails to be compatible with gradient memory reduction due to a contradiction between preserving gradients and releasing gradients. To address this issue, we propose a novel optimizer accumulation method for Adam, named Adam Accumulation (AdamA), which enables reducing both activation and gradient memory. Specifically, AdamA directly integrates gradients into optimizer states and accumulates optimizer states over micro-batches, so that gradients can be released immediately after use. We mathematically and experimentally demonstrate AdamA yields the same convergence properties as Adam. Evaluated on transformer-based models, AdamA achieves up to 23% memory reduction compared to gradient accumulation with less than 2% degradation in training throughput. Notably, AdamA can work together with memory reduction methods for optimizer states to fit 1.26x~3.14x larger models over PyTorch and DeepSpeed baseline on GPUs with different memory capacities. | Reject |
ICLR.cc/2021/Conference | Frequency Decomposition in Neural Processes | Neural Processes are a powerful tool for learning representations of function spaces purely from examples, in a way that allows them to perform predictions at test time conditioned on so-called context observations. The learned representations are finite-dimensional, while function spaces are infinite-dimensional, and so far it has been unclear how these representations are learned and what kinds of functions can be represented. We show that deterministic Neural Processes implicitly perform a decomposition of the training signals into different frequency components, similar to a Fourier transform. In this context, we derive a theoretical upper bound on the maximum frequency Neural Processes can reproduce, depending on their representation size. This bound is confirmed empirically. Finally, we show that Neural Processes can be trained to only represent a subset of possible frequencies and suppress others, which makes them programmable band-pass or band-stop filters. | Reject |
ICLR.cc/2020/Conference | Convolutional Conditional Neural Processes | We introduce the Convolutional Conditional Neural Process (ConvCNP), a new member of the Neural Process family that models translation equivariance in the data. Translation equivariance is an important inductive bias for many learning problems including time series modelling, spatial data, and images. The model embeds data sets into an infinite-dimensional function space, as opposed to finite-dimensional vector spaces. To formalize this notion, we extend the theory of neural representations of sets to include functional representations, and demonstrate that any translation-equivariant embedding can be represented using a convolutional deep-set. We evaluate ConvCNPs in several settings, demonstrating that they achieve state-of-the-art performance compared to existing NPs. We demonstrate that building in translation equivariance enables zero-shot generalization to challenging, out-of-domain tasks. | Accept (Talk) |
ICLR.cc/2022/Conference | LMSA: Low-relation Mutil-head Self-Attention Mechanism in Visual Transformer | The Transformer backbone network with the self-attention mechanism as the core has achieved great success in the field of natural language processing and computer vision. However, through the self-attention mechanism brings high performance, it also brings higher computational complexity compared to the classic visual feature extraction methods. To further reduce the complexity of self-attention mechanism and explore its lighter version in computer vision, in this paper, we design a novel lightweighted self-attention mechanism: Low-relation Mutil-head Self-Attention (LMSA), which is superior than the recent self-attention. Specifically, the proposed self-attention mechanism breaks the barrier of the dimensional consistency of the traditional self-attention mechanism, resulting in lower computational complexity and occupies less storage space. In addition, employing the new mechanism can release part of the computing consumption of the Transformer network and make the best use of it. Experimental results show that the dimensional consistency inside the traditional self-attention mechanism is unnecessary. In particular, using Swin as the backbone model for training, the accuracy in CIFAR-10 image classification task is improved by 0.43$\%$, in the meanwhile, the consumption of a single self-attention resource is reduced by 64.58$\%$, and the number of model parameters and model size are reduced by more than 15$\%$. By appropriately compressing the dimensions of the self-attention relationship variables, the Transformer network can be more efficient and even perform better. The results prompt us to rethink the reason why the self-attention mechanism works. | Reject |
ICLR.cc/2022/Conference | Sparse Fuse Dense: Towards High Quality 3D Detection With Depth Completion | Current LiDAR-only 3D detection methods inevitably suffer from the sparsity of point clouds. Sparse point clouds can confuse detectors as they lack sufficient geometric and semantic information. Many multi-modal methods are proposed to alleviate this issue, while different representations of images and point clouds make it difficult to fuse them, resulting in suboptimal performance. In this paper, we present a new multi-modal framework named SFD (Sparse Fuse Dense) to tackle these issues. Specifically, we propose to enhance sparse point clouds generated from LiDAR with dense pseudo point clouds generated from depth completion. To make full use of information from different types of point clouds, we design a new RoI feature fusion method 3D-GAF (3D Grid-wise Attentive Fusion), which fuses 3D RoI features from the couple of point clouds in a grid-wise attentive way. In addition, we devise a CPFE (Color Point Feature Extractor) to extract both 3D geometric and 2D semantic features in pseudo point clouds. Moreover, we introduce a multi-modal data augmentation method named SynAugment to utilize all data augmentation approaches tailored to LiDAR-only methods. Our method holds the highest entry on the KITTI 3D object detection leaderboard∗, demonstrating the effectiveness of SFD. Codes will be public. | Withdrawn |