Daily TMLR digest for Dec 23, 2023
TMLR
New certifications
==================
Survey Certification: Modular Deep Learning
Jonas Pfeiffer, Sebastian Ruder, Ivan Vulić, Edoardo Ponti
https://openreview.net/forum?id=z9EkXfvxta
---
Survey Certification: Benchmarks for Physical Reasoning AI
Andrew Melnik, Robin Schiewer, Moritz Lange, Andrei Ioan Muresanu, mozhgan saeidi, Animesh Garg, Helge Ritter
https://openreview.net/forum?id=cHroS8VIyN
---
Reproducibility Certification: StarCoder: may the source be with you!
Raymond Li, Loubna Ben allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia LI, Jenny Chim, Qian Liu, Evgenii Zheltonozhskii, Terry Yue Zhuo, Thomas Wang, Olivier Dehaene, Joel Lamy-Poirier, Joao Monteiro, Nicolas Gontier, Ming-Ho Yee, Logesh Kumar Umapathi, Jian Zhu, Ben Lipkin, Muhtasham Oblokulov, Zhiruo Wang, Rudra Murthy, Jason T Stillerman, Siva Sankalp Patel, Dmitry Abulkhanov, Marco Zocca, Manan Dey, Zhihan Zhang, Urvashi Bhattacharyya, Wenhao Yu, Sasha Luccioni, Paulo Villegas, Fedor Zhdanov, Tony Lee, Nadav Timor, Jennifer Ding, Claire S Schlesinger, Hailey Schoelkopf, Jan Ebert, Tri Dao, Mayank Mishra, Alex Gu, Carolyn Jane Anderson, Brendan Dolan-Gavitt, Danish Contractor, Siva Reddy, Daniel Fried, Dzmitry Bahdanau, Yacine Jernite, Carlos Muñoz Ferrandis, Sean Hughes, Thomas Wolf, Arjun Guha, Leandro Von Werra, Harm de Vries
https://openreview.net/forum?id=KoFOg41haE
---
Expert Certification: SHAP-XRT: The Shapley Value Meets Conditional Independence Testing
Jacopo Teneggi, Beepul Bharti, Yaniv Romano, Jeremias Sulam
https://openreview.net/forum?id=WFtTpQ47A7
---
Accepted papers
===============
Title: Resmax: An Alternative Soft-Greedy Operator for Reinforcement Learning
Authors: Erfan Miahi, Revan MacQueen, Alex Ayoub, Abbas Masoumzadeh, Martha White
Abstract: Soft-greedy operators, namely $\varepsilon$-greedy and softmax, remain a common choice to induce a basic level of exploration for action-value methods in reinforcement learning. These operators, however, have a few critical limitations. In this work, we investigate a simple soft-greedy operator, which we call resmax, that takes actions proportionally to their max action gap: the residual to the estimated maximal value. It is simple to use and ensures coverage of the state-space like $\varepsilon$-greedy, but focuses exploration more on potentially promising actions like softmax. Further, it does not concentrate probability as quickly as softmax, and so better avoids overemphasizing sub-optimal actions that appear high-valued during learning. Additionally, we prove it is a non-expansion for any fixed exploration hyperparameter, unlike the softmax policy which requires a state-action specific temperature to obtain a non-expansion (called mellowmax). We empirically validate that resmax is comparable to or outperforms $\varepsilon$-greedy and softmax across a variety of environments in tabular and deep RL.
URL: https://openreview.net/forum?id=wzzrs5QH5k
---
Title: Privacy Budget Tailoring in Private Data Analysis
Authors: Daniel Alabi, Chris Wiggins
Abstract: We consider the problem of learning differentially private linear and logistic regression models that do not exhibit disparate performance for minority groups in the data. Small-sized datasets pose a challenging regime for differential privacy; that is, satisfying differential privacy while learning models from data can lead to models with worse accuracy for minority---in size---subgroups. To address this challenge, inspired by Abowd & Schmutte (2018), we propose: (i) to systematically tailor the privacy budget to the different groups, (ii) use linear optimization oracles in a grid to optimize Lagrangian objectives that correspond to fair learning and optimization. We present efficient differentially private algorithms for linear and logistic regression subject to fairness constraints (e.g., bounded group loss) that allocate the privacy budget based on the private standard error of each subgroup in the data. Consequently, the formulation reduces the amount of noise added to these groups, which leads to more accurate models for such groups. We validate the proposed, group-aware budget allocation, method on synthetic and real-world datasets where we show significant reductions in prediction error for the smallest groups, while still preserving sufficient privacy to protect the minority group from re-identification attacks. In addition, we provide sample complexity lower bounds for our problem formulation.
URL: https://openreview.net/forum?id=SnPEhMyuYX
---
Title: Modular Deep Learning
Authors: Jonas Pfeiffer, Sebastian Ruder, Ivan Vulić, Edoardo Ponti
Abstract: Transfer learning has recently become the dominant paradigm of machine learning. Pre-trained models fine-tuned for downstream tasks achieve better performance with fewer labelled examples. Nonetheless, it remains unclear how to develop models that specialise towards multiple tasks without incurring negative interference and that generalise systematically to non-identically distributed tasks. Modular deep learning has emerged as a promising solution to these challenges. In this framework, units of computation are often implemented as autonomous parameter-efficient modules. Information is conditionally routed to a subset of modules and subsequently aggregated. These properties enable positive transfer and systematic generalisation by separating computation from routing and updating modules locally. We offer a survey of modular architectures, providing a unified view over several threads of research that evolved independently in the scientific literature. Moreover, we explore various additional purposes of modularity, including scaling language models, causal inference and discovery, programme simulation, and hierarchical reinforcement learning. Finally, we report various concrete applications where modularity has been successfully deployed such as cross-lingual and cross-modal knowledge transfer.
URL: https://openreview.net/forum?id=z9EkXfvxta
---
Title: Smoothed Differential Privacy
Authors: Ao Liu, Yu-Xiang Wang, Lirong Xia
Abstract: Differential privacy (DP) is a widely-accepted and widely-applied notion of privacy based on worst-case analysis. Often, DP classifies most mechanisms without additive noise as non-private (Dwork et al., 2014). Thus, additive noises are added to improve privacy (to achieve DP). However, in many real-world applications, adding additive noise is undesirable (Bagdasaryan et al., 2019) and sometimes prohibited (Liu et al., 2020).
In this paper, we propose a natural extension of DP following the worst average-case idea behind the celebrated smoothed analysis (Spielman & Teng, May 2004). Our notion, smoothed DP, can effectively measure the privacy leakage of mechanisms without additive noises under realistic settings. We prove that any discrete mechanism with sampling procedures is more private than what DP predicts, while many continuous mechanisms with sampling procedures are still non-private under smoothed DP. In addition, we prove several desirable properties of smoothed DP, including composition, robustness to post-processing, and distribution reduction. Based on those properties, we propose an efficient algorithm to calculate the privacy parameters for smoothed DP. Experimentally, we verify that, according to smoothed DP, the discrete sampling mechanisms are private in real-world elections, and some discrete neural networks can be private without adding any additive noise. We believe that these results contribute to the theoretical foundation of realistic privacy measures beyond worst-case analysis.
URL: https://openreview.net/forum?id=CviCLt44Em
---
Title: Distributed Architecture Search Over Heterogeneous Distributions
Authors: Erum Mushtaq, Chaoyang He, Jie Ding, Salman Avestimehr
Abstract: Federated learning (FL) is an efficient learning framework that assists distributed machine learning when data cannot be shared with a centralized server. Recent advancements in FL use predefined architecture-based learning for all clients. However, given that clients' data are invisible to the server and data distributions are non-identical across clients, a predefined architecture discovered in a centralized setting may not be an optimal solution for all the clients in FL. Motivated by this challenge, we introduce SPIDER, an algorithmic framework that aims to Search PersonalIzed neural architecture for feDERated learning. SPIDER is designed based on two unique features: (1) alternately optimizing one architecture-homogeneous global model in a generic FL manner and architecture-heterogeneous local models that are connected to the global model by weight-sharing-based regularization, (2) achieving architecture-heterogeneous local models by a perturbation-based neural architecture search method. Experimental results demonstrate superior prediction performance compared with other state-of-the-art personalization methods.
URL: https://openreview.net/forum?id=sY75NqDRk1
---
Title: DreamEdit: Subject-driven Image Editing
Authors: Tianle Li, Max Ku, Cong Wei, Wenhu Chen
Abstract: Subject-driven image generation aims at generating images containing customized subjects, which has recently drawn enormous attention from the research community. Nevertheless, the previous works cannot precisely control the background and position of the target subject. In this work, we aspire to fill the void of the existing subject-driven generation tasks. To this end, we propose two novel subject-driven editing sub-tasks, i.e., Subject Replacement and Subject Addition. The new tasks are challenging in multiple aspects: replacing a subject with a customized one can totally change its shape, texture, and color, while adding a target subject to a designated position in a provided scene necessitates a rational context-aware posture of the subject. To conquer these two novel tasks, we first manually curate a new dataset called DreamEditBench containing 22 different types of subjects, and 440 source images, which cover diverse scenarios with different difficulty levels. We plan to host DreamEditBench as a platform and hire trained evaluators for standardized human evaluation. We also devise an innovative method DreamEditor to resolve these tasks by performing iterative generation, which enables a smooth adaptation to the customized subject. In this project, we conduct automatic and human evaluations to understand the performance of our DreamEditor and baselines on DreamEditBench. We found that the new tasks are challenging for the existing models. For Subject Replacement, we found that the existing models are particularly sensitive to the shape and color of the original subject. When the original subject and the customized subject are highly different, the model failure rate will dramatically increase. For Subject Addition, we found that the existing models cannot easily blend the customized subjects into the background smoothly, which causes noticeable artifacts in the generated image. We hope that DreamEditBench can become a standardized platform to enable future investigations towards building more controllable subject-driven image editing. Our project and benchmark homepage is https://dreameditbenchteam.github.io/
URL: https://openreview.net/forum?id=P9haooN9v2
---
Title: UnIVAL: Unified Model for Image, Video, Audio and Language Tasks
Authors: Mustafa Shukor, Corentin Dancette, Alexandre Rame, Matthieu Cord
Abstract: Large Language Models (LLMs) have made the ambitious quest for generalist agents significantly far from being a fantasy. A key hurdle for building such general models is the diversity and heterogeneity of tasks and modalities. A promising solution is unification, allowing the support of a myriad of tasks and modalities within one unified framework. While few large models (e.g., Flamingo (Alayrac et al. 2022)), trained on massive datasets, can support more than two modalities, current small to mid-scale unified models are still limited to 2 modalities, usually image-text or video-text. The question that we ask is: is it possible to build efficiently a unified model that can support all modalities? To answer this, we propose UnIVAL, a step further towards this ambitious goal. Without relying on fancy datasets sizes or models with billions of parameters, the ~ 0.25B parameter UnIVAL model goes beyond two modalities and unifies text, images, video, and audio into a single model. Our model is efficiently pretrained on many tasks, based on task balancing and multimodal curriculum learning. UnIVAL shows competitive performance to existing state-of-the-art approaches, across image and video-text tasks. The feature representations learned from image and video-text modalities, allows the model to achieve competitive performance when finetuned on audio-text tasks, despite not being pretrained on audio. Thanks to the unified model, we propose a novel study on multimodal model merging via weight interpolation of models trained on different multimodal tasks, showing their benefits in particular for out-of-distribution generalization. Finally, we motivate unification by showing the synergy between tasks. The model weights and code are available at: https://github.com/mshukor/UnIVAL.
URL: https://openreview.net/forum?id=4uflhObpcp
---
Title: IndicTrans2: Towards High-Quality and Accessible Machine Translation Models for all 22 Scheduled Indian Languages
Authors: Jay Gala, Pranjal A Chitale, A K Raghavan, Varun Gumma, Sumanth Doddapaneni, Aswanth Kumar M, Janki Atul Nawale, Anupama Sujatha, Ratish Puduppully, Vivek Raghavan, Pratyush Kumar, Mitesh M Khapra, Raj Dabre, Anoop Kunchukuttan
Abstract: India has a rich linguistic landscape, with languages from 4 major language families spoken by over a billion people. 22 of these languages listed in the Constitution of India (referred to as scheduled languages) are the focus of this work. Given the linguistic diversity, high-quality and accessible Machine Translation (MT) systems are essential in a country like India. Before this work, there was (i) no parallel training data spanning all 22 languages, (ii) no robust benchmarks covering all these languages and containing content relevant to India, and (iii) no existing translation models that support all 22 scheduled languages of India. In this work, we aim to address this gap by focusing on the missing pieces required for enabling wide, easy, and open access to good machine translation systems for all 22 scheduled Indian languages. We identify four key areas of improvement: curating and creating larger training datasets, creating diverse and high-quality benchmarks, training multilingual models, and releasing models with open access. Our first contribution is the release of the Bharat Parallel Corpus Collection (BPCC), the largest publicly available parallel corpora for Indic languages. BPCC contains a total of 230M bitext pairs, of which a total of 126M were newly added, including 644K manually translated sentence pairs created as part of this work. Our second contribution is the release of the first $n$-way parallel benchmark covering all 22 Indian languages, featuring diverse domains, Indian-origin content, and conversational test sets. Next, we present IndicTrans2, the first translation model to support all 22 languages, surpassing existing models in performance on multiple existing and new benchmarks created as a part of this work. Lastly, to promote accessibility and collaboration, we release our models and associated data with permissive licenses at https://github.com/AI4Bharat/IndicTrans2.
URL: https://openreview.net/forum?id=vfT4YuzAYA
---
Title: Towards Optimization-Friendly Binary Neural Network
Authors: Nianhui Guo, Joseph Bethge, Hong Guo, Christoph Meinel, Haojin Yang
Abstract: Binary neural networks (BNNs) are a promising approach for compressing and accelerating deep learning models, especially in resource-constrained environments. However, the optimization gap between BNNs and their full-precision counterparts has long been an open problem limiting their performance. In this work, we propose a novel optimization pipeline to enhance the performance of BNNs. The main approach includes three key components: (1) BNext, a strong binary baseline based on an optimization-friendly basic block design, (2) knowledge complexity, a simple yet effective teacher-selection metric taking the capacity gap between teachers and binary students under consideration, (3) consecutive knowledge distillation (CKD), a novel multi-round optimization technique to transfer high-confidence knowledge from strong teachers to low-capacity BNNs.
We empirically validate the superiority of the method on several vision classification tasks CIFAR-10/100 & ImageNet. For instance, the BNext family outperforms previous BNNs under different capacity levels and contributes the first binary neural network to reach the state-of-the-art 80.57\% Top-1 accuracy on ImageNet with 0.82 GOPS, which verifies the potential of BNNs and already contributes a strong baseline for future research on high-accuracy BNNs. The code will be publicly available at (blind URL, see supplementary material).
URL: https://openreview.net/forum?id=4Hq816XDDG
---
Title: Equivariant MuZero
Authors: Andreea Deac, Theophane Weber, George Papamakarios
Abstract: Deep reinforcement learning has shown lots of success in closed, well-defined domains such as games (Chess, Go, StarCraft). The next frontier is real-world scenarios, where setups are numerous and varied. For this, agents need to learn the underlying environment dynamics, so as to robustly generalise to conditions that differ from those they were trained on. Model-based reinforcement learning algorithms, such as MuZero or Dreamer, aim to accomplish this by learning a world model. However, leveraging a world model has not yet consistently shown greater generalisation capabilities compared to model-free alternatives. In this work, we propose improving the data efficiency and generalisation capabilities of MuZero by explicitly incorporating the \emph{symmetries} of the environment in its world-model architecture. We prove that, so long as the neural networks used by MuZero are equivariant to a particular symmetry group acting on the environment, the entirety of MuZero's action-selection algorithm will also be equivariant to that group. As such, Equivariant MuZero is guaranteed to behave symmetrically in symmetrically-transformed states, and will hence be more data-efficient when learning its world models. We evaluate Equivariant MuZero on procedurally-generated MiniPacman and on Chaser from the ProcGen suite: training on a set of mazes, and then testing on unseen rotated versions, demonstrating the benefits of equivariance. We verify that our improvements hold even when only some of the components of Equivariant MuZero obey strict equivariance, which highlights the robustness of our construction.
URL: https://openreview.net/forum?id=ExbGarTbLE
---
Title: On the Efficacy of Differentially Private Few-shot Image Classification
Authors: Marlon Tobaben, Aliaksandra Shysheya, John F Bronskill, Andrew Paverd, Shruti Tople, Santiago Zanella-Beguelin, Richard E Turner, Antti Honkela
Abstract: There has been significant recent progress in training differentially private (DP) models which achieve accuracy that approaches the best non-private models. These DP models are typically pretrained on large public datasets and then fine-tuned on private downstream datasets that are relatively large and similar in distribution to the pretraining data. However, in many applications including personalization and federated learning, it is crucial to perform well (i) in the few-shot setting, as obtaining large amounts of labeled data may be problematic; and (ii) on datasets from a wide variety of domains for use in various specialist settings. To understand under which conditions few-shot DP can be effective, we perform an exhaustive set of experiments that reveals how the accuracy and vulnerability to attack of few-shot DP image classification models are affected as the number of shots per class, privacy level, model architecture, downstream dataset, and subset of learnable parameters in the model vary. We show that to achieve DP accuracy on par with non-private models, the shots per class must be increased as the privacy level increases. We also show that learning parameter-efficient FiLM adapters under DP is competitive with learning just the final classifier layer or learning all of the network parameters. Finally, we evaluate DP federated learning systems and establish state-of-the-art performance on the challenging FLAIR benchmark.
URL: https://openreview.net/forum?id=hFsr59Imzm
---
Title: Benchmarks for Physical Reasoning AI
Authors: Andrew Melnik, Robin Schiewer, Moritz Lange, Andrei Ioan Muresanu, mozhgan saeidi, Animesh Garg, Helge Ritter
Abstract: Physical reasoning is a crucial aspect in the development of general AI systems, given that human learning starts with interacting with the physical world before progressing to more complex concepts. Although researchers have studied and assessed the physical reasoning of AI approaches through various specific benchmarks, there is no comprehensive approach to evaluating and measuring progress. Therefore, we aim to offer an overview of existing benchmarks and their solution approaches and propose a unified perspective for measuring the physical reasoning capacity of AI systems. We select benchmarks that are designed to test algorithmic performance in physical reasoning tasks. While each of the selected benchmarks poses a unique challenge, their ensemble provides a comprehensive proving ground for an AI generalist agent with a measurable skill level for various physical reasoning concepts. This gives an advantage to such an ensemble of benchmarks over other holistic benchmarks that aim to simulate the real world by intertwining its complexity and many concepts. We group the presented set of physical reasoning benchmarks into subcategories so that more narrow generalist AI agents can be tested first on these groups.
URL: https://openreview.net/forum?id=cHroS8VIyN
---
Title: StarCoder: may the source be with you!
Authors: Raymond Li, Loubna Ben allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia LI, Jenny Chim, Qian Liu, Evgenii Zheltonozhskii, Terry Yue Zhuo, Thomas Wang, Olivier Dehaene, Joel Lamy-Poirier, Joao Monteiro, Nicolas Gontier, Ming-Ho Yee, Logesh Kumar Umapathi, Jian Zhu, Ben Lipkin, Muhtasham Oblokulov, Zhiruo Wang, Rudra Murthy, Jason T Stillerman, Siva Sankalp Patel, Dmitry Abulkhanov, Marco Zocca, Manan Dey, Zhihan Zhang, Urvashi Bhattacharyya, Wenhao Yu, Sasha Luccioni, Paulo Villegas, Fedor Zhdanov, Tony Lee, Nadav Timor, Jennifer Ding, Claire S Schlesinger, Hailey Schoelkopf, Jan Ebert, Tri Dao, Mayank Mishra, Alex Gu, Carolyn Jane Anderson, Brendan Dolan-Gavitt, Danish Contractor, Siva Reddy, Daniel Fried, Dzmitry Bahdanau, Yacine Jernite, Carlos Muñoz Ferrandis, Sean Hughes, Thomas Wolf, Arjun Guha, Leandro Von Werra, Harm de Vries
Abstract: The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license.
URL: https://openreview.net/forum?id=KoFOg41haE
---
Title: Mixture of Dynamical Variational Autoencoders for Multi-Source Trajectory Modeling and Separation
Authors: Xiaoyu Lin, Laurent Girin, Xavier Alameda-Pineda
Abstract: In this paper, we propose a latent-variable generative model called mixture of dynamical variational autoencoders (MixDVAE) to model the dynamics of a system composed of multiple moving sources. A DVAE model is pre-trained on a single-source dataset to capture the source dynamics. Then, multiple instances of the pre-trained DVAE model are integrated into a multi-source mixture model with a discrete observation-to-source assignment latent variable. The posterior distributions of both the discrete observation-to-source assignment variable and the continuous DVAE variables representing the sources content/position are estimated using the variational expectation-maximization algorithm, leading to multi-source trajectories estimation. We illustrate the versatility of the proposed MixDVAE model on two tasks: a computer vision task, namely multi-object tracking, and an audio processing task, namely single-channel audio source separation. Experimental results show that the proposed method works well on these two tasks, and outperforms several baseline methods.
URL: https://openreview.net/forum?id=sbkZKBVC31
---
Title: Fast Slate Policy Optimization: Going Beyond Plackett-Luce
Authors: Otmane Sakhi, David Rohde, Nicolas Chopin
Abstract: An increasingly important building block of large scale machine learning systems is based on returning slates; an ordered lists of items given a query. Applications of this technology include: search, information retrieval and recommender systems. When the action space is large, decision systems are restricted to a particular structure to complete online queries quickly. This paper addresses the optimization of these large scale decision systems given an arbitrary reward function. We cast this learning problem in a policy optimization framework and propose a new class of policies, born from a novel relaxation of decision functions. This results in a simple, yet efficient learning algorithm that scales to massive action spaces. We compare our method to the commonly adopted Plackett-Luce policy class and demonstrate the effectiveness of our approach on problems with action space sizes in the order of millions.
URL: https://openreview.net/forum?id=f7a8XCRtUu
---
Title: Error bounds and dynamics of bootstrapping in actor-critic reinforcement learning
Authors: Ahmed J Zerouali, Douglas Blair Tweed
Abstract: Actor-critic algorithms such as DDPG, TD3, and SAC, which are built on Silver's deterministic policy gradient theorem, are among the most successful reinforcement-learning methods, but their mathematical basis is not entirely clear. In particular, the critic networks in these algorithms learn to estimate action-value functions by a “bootstrapping” technique based on Bellman error, and it is unclear why this approach works so well in practice, given that Bellman error is only very loosely related to value error, i.e. to the inaccuracy of the action-value estimate. Here we show that policy training in this class of actor-critic methods depends not on the accuracy of the critic's action-value estimate but on how well the critic estimates the gradient of the action-value, which is better assessed using what we call difference error. We show that this difference error is closely related to the Bellman error — a finding which helps to explain why Bellman-based bootstrapping leads to good policies. Further, we show that value error and difference error show different dynamics along on-policy trajectories through state-action space: value error is a low-pass anticausal (i.e., backward-in-time) filter of Bellman error, and therefore accumulates along trajectories, whereas difference error is a high-pass filter of Bellman error. It follows that techniques which reduce the high-frequency Fourier components of the Bellman error may improve policy training even if they increase the actual size of the Bellman errors. These findings help to explain certain aspects of actor-critic methods that are otherwise theoretically puzzling, such as the use of policy (as distinct from exploratory) noise, and they suggest other measures that may improve these methods.
URL: https://openreview.net/forum?id=QCjMJfSnYk
---
Title: SHAP-XRT: The Shapley Value Meets Conditional Independence Testing
Authors: Jacopo Teneggi, Beepul Bharti, Yaniv Romano, Jeremias Sulam
Abstract: The complex nature of artificial neural networks raises concerns on their reliability, trustworthiness, and fairness in real-world scenarios. The Shapley value---a solution concept from game theory---is one of the most popular explanation methods for machine learning models. More traditionally, from a statistical perspective, feature importance is defined in terms of conditional independence. So far, these two approaches to interpretability and feature importance have been considered separate and distinct. In this work, we show that Shapley-based explanation methods and conditional independence testing are closely related. We introduce the \textbf{SHAP}ley E\textbf{X}planation \textbf{R}andomization \textbf{T}est (SHAP-XRT), a testing procedure inspired by the Conditional Randomization Test (CRT) for a specific notion of local (i.e., on a sample) conditional independence. With it, we prove that for binary classification problems, the marginal contributions in the Shapley value provide lower and upper bounds to the expected $p$-values of their respective tests. Furthermore, we show that the Shapley value itself provides an upper bound to the expected $p$-value of a global (i.e., overall) null hypothesis. As a result, we further our understanding of Shapley-based explanation methods from a novel perspective and characterize the conditions under which one can make statistically valid claims about feature importance via the Shapley value.
URL: https://openreview.net/forum?id=WFtTpQ47A7
---
Title: Federated Minimax Optimization with Client Heterogeneity
Authors: Pranay Sharma, Rohan Panda, Gauri Joshi
Abstract: Minimax optimization has seen a surge in interest with the advent of modern applications such as GANs, and it is inherently more challenging than simple minimization. The difficulty is exacerbated by the training data residing at multiple edge devices or \textit{clients}, especially when these clients can have heterogeneous datasets and heterogeneous local computation capabilities. We propose a general federated minimax optimization framework that subsumes such settings and several existing methods like Local SGDA. We show that naive aggregation of model updates made by clients running unequal number of local steps can result in optimizing a mismatched objective function -- a phenomenon previously observed in standard federated minimization. To fix this problem, we propose normalizing the client updates by the number of local steps. We analyze the convergence of the proposed algorithm for classes of nonconvex-concave and nonconvex-nonconcave functions and characterize the impact of heterogeneous client data, partial client participation, and heterogeneous local computations. For all the function classes considered, we significantly improve the existing computation and communication complexity results. Experimental results support our theoretical claims.
URL: https://openreview.net/forum?id=NnUmg1chLL
---
Title: Towards Fair Video Summarization
Authors: Anshuman Chhabra, Kartik Patwari, Chandana Kuntala, Sristi, Deepak Kumar Sharma, Prasant Mohapatra
Abstract: Automated video summarization is a vision task that aims to generate concise summaries of lengthy videos. Recent advancements in deep learning have led to highly performant video summarization models; however, there has been a lack of attention given to fairness and unbiased representation in the generated summaries. To bridge this gap, we introduce and analytically define the fair video summarization problem, and demonstrate its connections to the well-established problem of fair clustering. To facilitate fair model development, we also introduce the FairVidSum dataset, which is similar in design to state-of-the-art video summarization datasets such as TVSum and SumMe, but also includes annotations for sensitive attributes and individuals alongside frame importance scores. Finally, we propose the SumBal metric for quantifying the fairness of an outputted video summary. We conduct extensive experiments to benchmark the fairness of various state-of-the-art video summarization models. Our results highlight the need for better models that balance accuracy and fairness to ensure equitable representation and inclusion in summaries. For completeness, we also provide a novel fair-only baseline, FVS-LP, to showcase the fairness-utility gap models can improve upon.
URL: https://openreview.net/forum?id=Uj6MRfR1P5
---
New submissions
===============
Title: Layerwise complexity-matched learning yields an improved model of cortical area V2
Abstract: Human ability to recognize complex visual patterns arises through transformations performed by successive areas in the ventral visual cortex. Deep neural networks trained end-to-end for object recognition approach human capabilities, and offer the best descriptions to date of neural responses in the late stages of the hierarchy. But these networks provide a poor account of the early stages, compared to traditional hand-engineered models, or models optimized for coding efficiency or prediction. Moreover, the gradient backpropagation used in end-to-end learning is generally considered to be biologically implausible. Here, we overcome both of these limitations by developing a bottom-up self-supervised training methodology that operates independently on successive layers. Specifically, we maximize feature similarity between pairs of locally-deformed natural image patches, while decorrelating features across patches sampled from other images. Crucially, the deformation amplitudes are adjusted proportionally to receptive field sizes in each layer, thus matching the task complexity to the capacity at each stage of processing. In comparison with architecture-matched versions of previous models, we demonstrate that our layerwise complexity-matched learning (LCL) formulation produces a two-stage model (LCL-V2) that is better aligned with selectivity properties and neural activity in primate area V2. We demonstrate that the complexity-matched learning paradigm is critical for the emergence of the improved biological alignment. Finally, when the two-stage model is used as a fixed front-end for a deep network trained to perform object recognition, the resultant model (LCL-V2Net) is significantly better than standard end-to-end self-supervised, supervised, and adversarially-trained models in terms of generalization to out-of-distribution tasks and alignment with human behavior.
URL: https://openreview.net/forum?id=lQBsLfAWhj
---
Title: Improved Convergence of Score-Based Diffusion Models via Prediction-Correction
Abstract: Score-based generative models (SGMs) are powerful tools to sample from complex data distributions. Their underlying idea is to \emph{(i)} run a forward process for time $T_1$ by adding noise to the data, \emph{(ii)} estimate its score function, and \emph{(iii)} use such estimate to run a reverse process. As the reverse process is initialized with the stationary distribution of the forward one, the existing analysis paradigm requires $T_1\to\infty$. This is however problematic: from a theoretical viewpoint, for a given precision of the score approximation, the convergence guarantee fails as $T_1$ diverges; from a practical viewpoint, a large $T_1$ increases computational costs and leads to error propagation.
This paper addresses the issue by considering a version of the popular \emph{predictor-corrector} scheme: after running the forward process, we first estimate the final distribution via an inexact Langevin dynamics and then revert the process. Our key technical contribution is to provide convergence guarantees which require to run the forward process \emph{only for a fixed finite time} $T_1$.
Our bounds exhibit a mild logarithmic dependence on the input dimension and the subgaussian norm of the target distribution, have minimal assumptions on the data, and require only to control the $L^2$ loss on the score approximation, which is the quantity minimized in practice.
URL: https://openreview.net/forum?id=0zKvH7YiAq
---
Title: Path Development Network with Finite-dimensional Lie Group
Abstract: Signature, lying at the heart of rough path theory, is a central tool for analysing controlled differential equations driven by irregular paths. Recently it has also found extensive applications in machine learning and data science as a mathematically principled, universal feature that boosts the performance of deep learning-based models in sequential data tasks. It, nevertheless, suffers from the curse of dimensionality when paths are high-dimensional.
We propose a novel, trainable path development layer, which exploits representations of sequential data through finite-dimensional Lie groups, thus resulting in dimension reduction. Its backpropagation algorithm is designed via optimization on manifolds. Our proposed layer, analogous to recurrent neural networks (RNN), possesses an explicit, simple recurrent unit that alleviates the gradient issues.
Our layer demonstrates its strength in irregular time series modelling. Empirical results on a range of datasets show that the development layer consistently and significantly outperforms signature features on accuracy and dimensionality. The compact hybrid model (stacking one-layer LSTM with the development layer) achieves state-of-the-art against various RNN and continuous time series models. Our layer also enhances the performance of modelling dynamics constrained to Lie groups.
Code is available at \url{https://github.com/PDevNet/DevNet.git}.
URL: https://openreview.net/forum?id=5xggNifExF
---
Title: Using Higher-Order Moments to Assess the Quality of GAN-generated Image Features
Abstract: The rapid advancement of Generative Adversarial Networks (GANs) necessitates the need to robustly evaluate these models. Among the established evaluation criteria, the Fréchet Inception Distance (FID) has been widely adopted due to its conceptual simplicity, fast computation time, and strong correlation with human perception. However, FID has inherent limitations, mainly stemming from its assumption that feature embeddings follow a Gaussian distribution, and therefore can be defined by their first two moments. As this does not hold in practice, in this paper we explore the importance of third-moments in image feature data and use this information to define a new measure, which we call the Skew Inception Distance (SID). We prove that SID is a pseudometric on probability distributions, show how it extends FID, and present a practical method for its computation. Our numerical experiments support that SID either tracks with FID or, in some cases, aligns more closely with human perception when evaluating image features of ImageNet data. Our work also shows that principal component analysis can be used to speed up the computation time of both FID and SID. Although we focus on using SID on image features for GAN evaluation, SID is applicable much more generally, including for the evaluation of other generative models.
URL: https://openreview.net/forum?id=Io3jDUC4DP
---
Title: GUARD: A Safe Reinforcement Learning Benchmark
Abstract: Due to the trial-and-error nature, it is typically challenging to apply RL algorithms to safety-critical real-world applications, such as autonomous driving, human-robot interaction, robot manipulation, etc, where such errors are not tolerable. Recently, safe RL (i.e. constrained RL) has emerged rapidly in the literature, in which the agents explore the environment while satisfying constraints. Due to the diversity of algorithms and tasks, it remains difficult to compare existing safe RL algorithms. To fill that gap, we introduce GUARD, a Generalized Unified SAfe Reinforcement Learning Development Benchmark. GUARD has several advantages compared to existing benchmarks. First, GUARD is a generalized benchmark with a wide variety of RL agents, tasks, and safety constraint specifications. Second, GUARD comprehensively covers state-of-the-art safe RL algorithms with self-contained implementations. Third, GUARD is highly customizable in tasks and algorithms. We present a comparison of state-of-the-art on-policy safe RL algorithms in various task settings using GUARD and establish baselines that future work can build on.
URL: https://openreview.net/forum?id=kZFKwApeQO
---
Title: Accurate Neural Network Pruning Requires Rethinking Sparse Optimization
Abstract: Obtaining versions of deep neural networks that are both highly-accurate and highly-sparse is one of the
main challenges in the area of model compression, and several high-performance pruning techniques have been investigated by the community. Yet, much less is known about the interaction between sparsity and the standard stochastic optimization techniques used for training sparse networks, and most existing work uses standard dense schedules and hyperparameters for training sparse networks. In this work, we examine the impact of high sparsity on model training using the standard computer vision and natural language processing sparsity benchmarks. We begin by showing that using standard dense training recipes for sparse training is suboptimal, and results in under-training. We provide new approaches for mitigating this issue for both sparse pre-training of vision models (e.g. ResNet50/ImageNet) and sparse fine-tuning of language models (e.g. BERT/GLUE), achieving state-of-the-art results in both settings in the high-sparsity regime, and providing detailed analyses for the difficulty of sparse training in both scenarios. Our work sets a new threshold in terms of the accuracies that can be achieved under high sparsity, and should inspire further research into improving sparse model training, to reach higher accuracies under high sparsity, but also to do so efficiently.
URL: https://openreview.net/forum?id=vgthYeRBAF
---
Title: Towards interpretable-by-design deep learning algorithms
Abstract: Most of the existing deep learning (DL) methods rely on parametric tuning and lack explainability. The few methods that claim to offer explainable DL solutions, such as ProtoPNet and xDNN, require end-to-end training and finetuning. In this study, we propose a framework called IDEAL (Interpretable-by-design DEep learning ALgorithms) which recasts the standard supervised classification problem into a function of similarity to a set of prototypes derived from the training data, while taking advantage of existing latent spaces of large neural networks forming so-called Foundation Models (FM). Using the IDEAL approach we can decompose the overall problem into two inherently connected stages: A) feature extraction (FE), which maps the raw features of the real-world data into a latent space, and B) identification of representative prototypes and decision making based on similarity and association between the query and the prototypes. This addresses the issue of interpretability (stage B) while retaining the benefits from the tremendous achievements offered by DL models (e.g., visual transformers, ViT) which are often pre-trained on huge datasets such as IG-3.6B + ImageNet-1K or LVD-142M (stage A). The key findings can be summarized as follows: (1) the proposed models are interpretable through prototypes, while also mitigating the issue of confounding bias, (2) the IDEAL framework circumvents the issue of catastrophic forgetting, allowing efficient class-incremental learning, and (3) the IDEAL approach demonstrates that ViT architectures narrow the gap between finetuned and non-finetuned models allowing for transfer learning in a fraction of time without finetuning of the feature space on a target dataset with iterative supervised methods. Using a range of datasets (CIFAR-10, CIFAR-100, CalTech101, STL-10, Oxford-IIIT Pet, EuroSAT), we demonstrate, through an extensive set of experiments, how the choice of the latent space, prototype selection, and finetuning of the latent space affect the performance of the models. Building upon this knowledge, we demonstrate that the proposed models have an advantage over state-of-the-art baselines in class-incremental learning. Finally, we analyse the interpretations provided by the proposed IDEAL framework, as well as the impact of confounding on the interpretations, demonstrating that the proposed approach without finetuning improves the performance on confounded data over finetuned counterparts.
URL: https://openreview.net/forum?id=PRFe38d9HE
---
Title: Enhancing Vision-Language Model with Unmasked Token Alignment
Abstract: Contrastive pre-training on image-text pairs, exemplified by CLIP, becomes a standard technique for learning multi-modal visual-language representations. Although CLIP has demonstrated remarkable performance, training it from scratch on noisy web-scale datasets is computationally demanding. On the other hand, mask-then-predict pre-training approaches, like Masked Image Modeling (MIM), offer efficient self-supervised learning for single-modal representations. This paper introduces $\textbf{U}$nmasked $\textbf{T}$oken $\textbf{A}$lignment ($\textbf{UTA}$), a method that leverages existing CLIP models to further enhance its vision-language representations. UTA trains a Vision Transformer (ViT) by aligning unmasked visual tokens to the corresponding image tokens from a frozen CLIP vision encoder, which automatically aligns the ViT model with the CLIP text encoder. The pre-trained ViT can be directly applied for zero-shot evaluation even without training on image-text pairs. Compared to MIM approaches, UTA does not suffer from training-finetuning inconsistency and is much more training-efficient by avoiding using the extra $\mathrm{[MASK]}$ tokens. Extensive experimental results demonstrate that UTA can enhance CLIP models and outperform existing MIM methods on various uni- and multi-modal benchmarks.
URL: https://openreview.net/forum?id=JkFEVbW6wE
---
Title: Semantic Positive Pairs for Enhancing Visual Representation Learning of Instance Discrimination methods
Abstract: Self-supervised learning algorithms (SSL) based on instance discrimination have shown promising results, performing competitively or even outperforming supervised learning counterparts in some downstream tasks. Such approaches employ data augmentation to create
two views of the same instance (i.e., positive pairs) and encourage the model to learn good representations by attracting these views closer in the embedding space without collapsing to the trivial solution. However, data augmentation is limited in representing positive pairs, and the repulsion process between the instances during contrastive learning may discard important features for instances that have similar categories. To address this issue, we pro pose an approach to identify those images with similar semantic content and treat them as positive instances, thereby reducing the chance of discarding important features during representation learning and increasing the richness of the latent representation. Our approach is generic and could work with any self-supervised instance discrimination frameworks such as MoCo and SimSiam. To evaluate our method, we run experiments on three benchmark datasets: ImageNet, STL-10 and CIFAR-10 with different instance discrimination SSL approaches. The experimental results show that our approach consistently outperforms the baseline methods across all three datasets; for instance, we improve upon the vanilla MoCo-v2 by 4.1% on ImageNet under a linear evaluation protocol over 800 epochs.
URL: https://openreview.net/forum?id=z5AXLMBWdU
---
Title: How good is Good-Turing for Markov samples?
Abstract: The Good-Turing (GT) estimator for the missing mass (i.e., total probability of missing symbols) in $n$ samples is the number of symbols that appeared exactly once divided by $n$. For i.i.d. samples, the bias and squared-error risk of the GT estimator can be shown to fall as $1/n$ by bounding the expected error uniformly over all symbols. In this work, we study convergence of the GT estimator for missing stationary mass (i.e., total stationary probability of missing symbols) of Markov samples on an alphabet $\mathcal{X}$ with stationary distribution $[\pi_x:x\in\cX]$ and transition probability matrix (t.p.m.) $P$. This is an important and interesting problem because GT is widely used in applications with temporal dependencies such as language models assigning probabilities to word sequences, which are modelled as Markov. We show that convergence of GT depends on convergence of $(P^{\sim x})^n$, where $P^{\sim x}$ is $P$ with the $x$-th column zeroed out. This, in turn, depends on the Perron eigenvalue $\lambda^{\sim x}$ of $P^{\sim x}$ and its relationship with $\pi_x$ uniformly over $x$. For randomly generated t.p.ms and t.p.ms derived from New York Times and Charles Dickens corpora, we numerically exhibit such uniform-over-$x$ relationships between $\lambda^{\sim x}$ and $\pi_x$. This supports the observed success of GT in language models and practical text data scenarios. For Markov chains with rank-2, diagonalizable t.p.ms having spectral gap $\beta$, we show minimax rate upper and lower bounds of $1/(n\beta^5)$ and $1/(n\beta)$, respectively, for the estimation of stationary missing mass. This theoretical result extends the $1/n$ minimax rate for i.i.d. or rank-1 t.p.ms to rank-2 Markov, and is a first such minimax rate result for missing mass of Markov samples. We also show, through experiments, that the MSE of GT decays at a slower rate as the rank of the t.p.m increases.
URL: https://openreview.net/forum?id=KokkP2nQ24
---