Self-supervised learning methods have shown impressive results in downstream classification tasks. However, there is limited work in understanding their failure models and interpreting the learned representations of these models. In this paper, we tackle these issues and study the representation space of self-supervised models by understanding the underlying reasons for misclassifications in a downstream task. Over several state-of-the-art self-supervised models including SimCLR, SwaV, MoCo V2 and BYOL, we observe that representations of correctly classified samples have few discriminative features with highly deviated values compared to other features. This is in a clear contrast with representations of misclassified samples. We also observe that noisy features in the representation space often correspond to spurious attributes in images making the models less interpretable. Building on these observations, we propose a sample-wise Self-Supervised Representation Quality Score (or, Q-Score) that, without access to any label information, is able to predict if a given sample is likely to be misclassified in the downstream task, achieving an AUPRC of up to 0.90. Q-Score can also be used as a regularization to remedy low-quality representations leading to 3.26% relative improvement in accuracy of SimCLR on ImageNet-100. Moreover, we show that Q-Score regularization increases representation sparsity, thus reducing noise and improving interpretability through gradient heatmaps.
Using natural language as a supervision for training visual recognition models holds great promise. Recent works have shown that if such supervision is used in the form of alignment between images and captions in large training datasets, then the resulting aligned models perform well on zero-shot classification as downstream tasks2. In this paper, we focus on teasing out what parts of the language supervision are essential for training zero-shot image classification models. Through extensive and careful experiments, we show that: 1) A simple Bag-of-Words (BoW) caption could be used as a replacement for most of the image captions in the dataset. Surprisingly, we observe that this approach improves the zero-shot classification performance when combined with word balancing. 2) Using a BoW pretrained model, we can obtain more training data by generating pseudo-BoW captions on images that do not have a caption. Models trained on images with real and pseudo-BoW captions achieve stronger zero-shot performance. On ImageNet-1k zero-shot evaluation, our best model, that uses only 3M image-caption pairs, performs on-par with a CLIP model trained on 15M image-caption pairs (31.5% vs 31.3%).
While neural networks have shown remarkable success on classification tasks in terms of average-case performance, they often fail to perform well on certain groups of the data. Such group information may be expensive to obtain; thus, recent works in robustness and fairness have proposed ways to improve worst-group performance even when group labels are unavailable for the training data. However, these methods generally underperform methods that utilize group information at training time. In this work, we assume access to a small number of group labels alongside a larger dataset without group labels. We propose BARACK, a simple two-step framework to utilize this partial group information to improve worst-group performance: train a model to predict the missing group labels for the training data, and then use these predicted group labels in a robust optimization objective. Theoretically, we provide generalization bounds for our approach in terms of the worst-group performance, showing how the generalization error scales with respect to both the total number of training points and the number of training points with group labels. Empirically, our method outperforms the baselines that do not use group information, even when only 1-33% of points have group labels. We provide ablation studies to support the robustness and extensibility of our framework.
An extractive rationale explains a language model's (LM's) prediction on a given task instance by highlighting the text inputs that most influenced the output. Ideally, rationale extraction should be faithful (reflects LM's behavior), plausible (makes sense to humans), data-efficient, and fast, without sacrificing the LM's task performance. Prior rationale extraction works consist of specialized approaches for addressing various subsets of these desiderata -- but never all five. Narrowly focusing on certain desiderata typically comes at the expense of ignored ones, so existing rationale extractors are often impractical in real-world applications. To tackle this challenge, we propose UniREx, a unified and highly flexible learning framework for rationale extraction, which allows users to easily account for all five factors. UniREx enables end-to-end customization of the rationale extractor training process, supporting arbitrary: (1) heuristic/learned rationale extractors, (2) combinations of faithfulness and/or plausibility objectives, and (3) amounts of gold rationale supervision. Across three text classification datasets, our best UniREx configurations achieve a superior balance of the five desiderata, when compared to strong baselines. Furthermore, UniREx-trained rationale extractors can even generalize to unseen datasets and tasks.
Exponential tilting is a technique commonly used in fields such as statistics, probability, information theory, and optimization to create parametric distribution shifts. Despite its prevalence in related fields, tilting has not seen widespread use in machine learning. In this work, we aim to bridge this gap by exploring the use of tilting in risk minimization. We study a simple extension to ERM -- tilted empirical risk minimization (TERM) -- which uses exponential tilting to flexibly tune the impact of individual losses. The resulting framework has several useful properties: We show that TERM can increase or decrease the influence of outliers, respectively, to enable fairness or robustness; has variance-reduction properties that can benefit generalization; and can be viewed as a smooth approximation to a superquantile method. Our work makes rigorous connections between TERM and related objectives, such as Value-at-Risk, Conditional Value-at-Risk, and distributionally robust optimization (DRO). We develop batch and stochastic first-order optimization methods for solving TERM, provide convergence guarantees for the solvers, and show that the framework can be efficiently solved relative to common alternatives. Finally, we demonstrate that TERM can be used for a multitude of applications in machine learning, such as enforcing fairness between subgroups, mitigating the effect of outliers, and handling class imbalance. Despite the straightforward modification TERM makes to traditional ERM objectives, we find that the framework can consistently outperform ERM and deliver competitive performance with state-of-the-art, problem-specific approaches.
Large neural networks are impractical to deploy on mobile devices due to their heavy computational cost and slow inference. Knowledge distillation (KD) is a technique to reduce the model size while retaining performance by transferring knowledge from a large "teacher" model to a smaller "student" model. However, KD on multimodal datasets such as vision-language datasets is relatively unexplored and digesting such multimodal information is challenging since different modalities present different types of information. In this paper, we propose modality-specific distillation (MSD) to effectively transfer knowledge from a teacher on multimodal datasets. Existing KD approaches can be applied to multimodal setup, but a student doesn't have access to modality-specific predictions. Our idea aims at mimicking a teacher's modality-specific predictions by introducing an auxiliary loss term for each modality. Because each modality has different importance for predictions, we also propose weighting approaches for the auxiliary losses; a meta-learning approach to learn the optimal weights on these loss terms. In our experiments, we demonstrate the effectiveness of our MSD and the weighting scheme and show that it achieves better performance than KD.
Quantization of the parameters of machine learning models, such as deep neural networks, requires solving constrained optimization problems, where the constraint set is formed by the Cartesian product of many simple discrete sets. For such optimization problems, we study the performance of the Alternating Direction Method of Multipliers for Quantization ($\texttt{ADMM-Q}$) algorithm, which is a variant of the widely-used ADMM method applied to our discrete optimization problem. We establish the convergence of the iterates of $\texttt{ADMM-Q}$ to certain $\textit{stationary points}$. To the best of our knowledge, this is the first analysis of an ADMM-type method for problems with discrete variables/constraints. Based on our theoretical insights, we develop a few variants of $\texttt{ADMM-Q}$ that can handle inexact update rules, and have improved performance via the use of "soft projection" and "injecting randomness to the algorithm". We empirically evaluate the efficacy of our proposed approaches.
Empirical risk minimization (ERM) is typically designed to perform well on the average loss, which can result in estimators that are sensitive to outliers, generalize poorly, or treat subgroups unfairly. While many methods aim to address these problems individually, in this work, we explore them through a unified framework---tilted empirical risk minimization (TERM). In particular, we show that it is possible to flexibly tune the impact of individual losses through a straightforward extension to ERM using a hyperparameter called the tilt. We provide several interpretations of the resulting framework: We show that TERM can increase or decrease the influence of outliers, respectively, to enable fairness or robustness; has variance-reduction properties that can benefit generalization; and can be viewed as a smooth approximation to a superquantile method. We develop batch and stochastic first-order optimization methods for solving TERM, and show that the problem can be efficiently solved relative to common alternatives. Finally, we demonstrate that TERM can be used for a multitude of applications, such as enforcing fairness between subgroups, mitigating the effect of outliers, and handling class imbalance. TERM is not only competitive with existing solutions tailored to these individual problems, but can also enable entirely new applications, such as simultaneously addressing outliers and promoting fairness.
Federated learning aims to jointly learn statistical models over massively distributed remote devices. In this work, we propose FedDANE, an optimization method that we adapt from DANE, a method for classical distributed optimization, to handle the practical constraints of federated learning. We provide convergence guarantees for this method when learning over both convex and non-convex functions. Despite encouraging theoretical results, we find that the method has underwhelming performance empirically. In particular, through empirical simulations on both synthetic and real-world datasets, FedDANE consistently underperforms baselines of FedAvg and FedProx in realistic federated settings. We identify low device participation and statistical device heterogeneity as two underlying causes of this underwhelming performance, and conclude by suggesting several directions of future work.
We study the optimization problem for decomposing $d$ dimensional fourth-order Tensors with $k$ non-orthogonal components. We derive \textit{deterministic} conditions under which such a problem does not have spurious local minima. In particular, we show that if $\kappa = \frac{\lambda_{max}}{\lambda_{min}} < \frac{5}{4}$, and incoherence coefficient is of the order $O(\frac{1}{\sqrt{d}})$, then all the local minima are globally optimal. Using standard techniques, these conditions could be easily transformed into conditions that would hold with high probability in high dimensions when the components are generated randomly. Finally, we prove that the tensor power method with deflation and restarts could efficiently extract all the components within a tolerance level $O(\kappa \sqrt{k\tau^3})$ that seems to be the noise floor of non-orthogonal tensor decomposition.