ICML 2024 Papers

Semi-supervised learning (SSL) is a classical machine learning paradigm dealing with labeled and unlabeled data. However, it often suffers performance degradation in real-world open-set scenarios, where unlabeled data contains outliers from novel categories that do not appear in labeled data. Existing studies commonly tackle this challenging open-set SSL problem with detect-and-filter strategy, which attempts to purify unlabeled data by detecting and filtering outliers. In this paper, we propose a novel binary decomposition strategy, which refrains from error-prone procedure of outlier detection by directly transforming the original open-set SSL problem into a number of standard binary SSL problems. Accordingly, a concise yet effective approach named BDMatch is presented. BDMatch confronts two attendant issues brought by binary decomposition, i.e. class-imbalance and representation-compromise, with adaptive logit adjustment and label-specific feature learning respectively. Comprehensive experiments on diversified benchmarks clearly validate the superiority of BDMatch as well as the effectiveness of our binary decomposition strategy.

The ability of deep networks to learn superior representations hinges on leveraging the proper inductive biases, considering the inherent properties of datasets. In tabular domains, it is critical to effectively handle heterogeneous features (both categorical and numerical) in a unified manner and to grasp irregular functions like piecewise constant functions. To address the challenges in the self-supervised learning framework, we propose a novel pretext task based on the classical binning method. The idea is straightforward: reconstructing the bin indices (either orders or classes) rather than the original values. This pretext task provides the encoder with an inductive bias to capture the irregular dependencies, mapping from continuous inputs to discretized bins, and mitigates the feature heterogeneity by setting all features to have category-type targets. Our empirical investigations ascertain several advantages of binning: capturing the irregular function, compatibility with encoder architecture and additional modifications, standardizing all features into equal sets, grouping similar values within a feature, and providing ordering information. Comprehensive evaluations across diverse tabular datasets corroborate that our method consistently improves tabular representation learning performance for a wide range of downstream tasks. The codes are available in https://github.com/kyungeun-lee/tabularbinning.

This paper introduces a novel approach to improving the training stability of self-supervised learning (SSL) methods by leveraging a non-parametric memory of seen concepts. The proposed method involves augmenting a neural network with a memory component to stochastically compare current image views with previously encountered concepts. Additionally, we introduce stochastic memory blocks to regularize training and enforce consistency between image views. We extensively benchmark our method on many vision tasks, such as linear probing, transfer learning, few-shot classification, and image retrieval on many datasets. The experimental results consolidate the effectiveness of the proposed approach in achieving stable SSL training without additional regularizers while learning highly transferable representations and requiring less computing time and resources.

Self-attention has been widely used in various machine learning models, such as vision transformers. The standard dot-product self-attention is arguably the most popular structure, and there is a growing interest in understanding the mathematical properties of such attention mechanisms. This paper presents a fine-grained local sensitivity analysis of the standard dot-product self-attention, leading to new non-vacuous certified robustness results for vision transformers. Despite the well-known fact that dot-product self-attention is not (globally) Lipschitz, we develop new theoretical analysis of Local Fine-grained Attention Sensitivity (LoFAST) quantifying the effect of input feature perturbations on the attention output. Our analysis reveals that the local sensitivity of dot-product self-attention to $\ell_2$ perturbations can actually be controlled by several key quantities associated with the attention weight matrices and the unperturbed input. We empirically validate our theoretical findings by computing non-vacuous certified $\ell_2$-robustness for vision transformers on CIFAR-10 and SVHN datasets. The code for LoFAST is available at https://github.com/AaronHavens/LoFAST.

The strong performance of neural networks typically hinges on the availability of extensive labeled data, yet acquiring ground-truth labels is often challenging. Instead, noisy supervisions from multiple sources, e.g., by multiple well-designed rules, are more convenient to collect. In this paper, we focus on the realistic problem of learning from multiple noisy label sources, and argue that prior studies have overlooked the crucial *self-cognition* ability of neural networks, i.e., the inherent capability of autonomously distinguishing noise during training. We theoretically analyze this ability of neural networks when meeting multiple noisy label sources, which reveals that neural networks possess the capability to recognize both instance-wise noise within each single noisy label source and annotator-wise quality among multiple noisy label sources. Inspired by the theoretical analyses, we introduce an approach named Self-cognitive Denoising for Multiple noisy label sources (SDM), which exploits the self-cognition ability of neural networks to denoise during training. Furthermore, we build a selective distillation module following the theoretical insights to optimize computational efficiency. The experiments on various datasets demonstrate the superiority of our method.

We introduce a novel Pseudo-Negative Regularization (PNR) framework for effective continual self-supervised learning (CSSL). Our PNR leverages pseudo-negatives obtained through model-based augmentation in a way that newly learned representations may not contradict what has been learned in the past. Specifically, for the InfoNCE-based contrastive learning methods, we define symmetric pseudo-negatives obtained from current and previous models and use them in both main and regularization loss terms. Furthermore, we extend this idea to non-contrastive learning methods which do not inherently rely on negatives. For these methods, a pseudo-negative is defined as the output from the previous model for a differently augmented version of the anchor sample and is asymmetrically applied to the regularization term. Extensive experimental results demonstrate that our PNR framework achieves state-of-the-art performance in representation learning during CSSL by effectively balancing the trade-off between plasticity and stability.

Popular prompt strategies like Chain-of-Thought Prompting can dramatically improve the reasoning abilities of Large Language Models (LLMs) in various domains. However, such hand-crafted prompt-strategies are often sub-optimal. In this paper, we present Promptbreeder, a general-purpose self-referential self-improvement mechanism that evolves and adapts prompts for a given domain. Driven by an LLM, Promptbreeder mutates a population of task-prompts, evaluates them for fitness on a training set, and repeats this process over multiple generations to evolve task-prompts. Crucially, the mutation of these task-prompts is governed by mutation-prompts that the LLM generates and improves throughout evolution in a self-referential way. That is, Promptbreeder is not just improving task-prompts, but it is also improving the mutation-prompts that improve these task-prompts. Promptbreeder outperforms state-of-the-art prompt strategies such as Chain-of-Thought and Plan-and-Solve Prompting on commonly used arithmetic and commonsense reasoning benchmarks. Furthermore, Promptbreeder is able to evolve intricate task-prompts for the challenging problem of hate speech classification.

Modern computer vision offers a great variety of models to practitioners, and selecting a model from multiple options for specific applications can be challenging. Conventionally, competing model architectures and training protocols are compared by their classification accuracy on ImageNet. However, this single metric does not fully capture performance nuances critical for specialized tasks. In this work, we conduct an in-depth comparative analysis of model behaviors beyond ImageNet accuracy, for both ConvNet and Vision Transformer architectures, each across supervised and CLIP training paradigms. Although our selected models have similar ImageNet accuracies and compute requirements, we find that they differ in many other aspects: types of mistakes, output calibration, transferability, and feature invariance, among others. This diversity in model characteristics, not captured by traditional metrics, highlights the need for more nuanced analysis when choosing among different models.

We delve into the challenge of semi-supervised node classification on the Contextual Stochastic Block Model (CSBM) dataset. Here, nodes from the two-cluster Stochastic Block Model (SBM) are coupled with feature vectors, which are derived from a Gaussian Mixture Model (GMM) that corresponds to their respective node labels. With only a subset of the CSBM node labels accessible for training, our primary objective becomes the accurate classification of the remaining nodes. Venturing into the transductive learning landscape, we, for the first time, pinpoint the information-theoretical threshold for the exact recovery of all test nodes in CSBM. Concurrently, we design an optimal spectral estimator inspired by Principal Component Analysis (PCA) with the training labels and essential data from both the adjacency matrix and feature vectors. We also evaluate the efficacy of graph ridge regression and Graph Convolutional Networks (GCN) on this synthetic dataset. Our findings underscore that graph ridge regression and GCN possess the ability to achieve the information threshold of exact recovery in a manner akin to the optimal estimator when using the optimal weighted self-loops. This highlights the potential role of feature learning in augmenting the proficiency of GCN, especially in the realm of semi-supervised learning.

To interact with humans and act in the world, agents need to understand the range of language that people use and relate it to the visual world. While current agents can learn to execute simple language instructions, we aim to build agents that leverage diverse language---language like "this button turns on the TV" or "I put the bowls away"---that conveys general knowledge, describes the state of the world, provides interactive feedback, and more. Our key idea is that *agents should interpret such diverse language as a signal that helps them predict the future*: what they will observe, how the world will behave, and which situations will be rewarded. This perspective unifies language understanding with future prediction as a powerful self-supervised learning objective. We instantiate this in Dynalang, an agent that learns a multimodal world model to predict future text and image representations, and learns to act from imagined model rollouts. While current methods that learn language-conditioned policies degrade in performance with more diverse types of language, we show that Dynalang learns to leverage environment descriptions, game rules, and instructions to excel on tasks ranging from game-playing to navigating photorealistic home scans. Finally, we show that our method enables additional capabilities due to learning a generative model: Dynalang can be pretrained on text-only data, enabling learning from offline datasets, and generate language grounded in an environment.

Modern language models rely on the transformer architecture and attention mechanism to perform language understanding and text generation. In this work, we study learning a 1-layer self-attention model from a set of prompts and the associated outputs sampled from the model. We first establish a formal link between the self-attention mechanism and Markov models under suitable conditions: Inputting a prompt to the self-attention model samples the output token according to a *context-conditioned Markov chain* (CCMC). *CCMC* is obtained by weighing the transition matrix of a standard Markov chain according to the sufficient statistics of the prompt/context. Building on this formalism, we develop identifiability/coverage conditions for the data distribution that guarantee consistent estimation of the latent model under a teacher-student setting and establish sample complexity guarantees under IID data. Finally, we study the problem of learning from a single output trajectory generated in response to an initial prompt. We characterize a *winner-takes-all* phenomenon where the generative process of self-attention evolves to sampling from a small set of *winner tokens* that dominate the context window. This provides a mathematical explanation to the tendency of modern LLMs to generate repetitive text.

We propose a method to improve the efficiency and accuracy of amortized Bayesian inference by leveraging universal symmetries in the joint probabilistic model of parameters and data. In a nutshell, we invert Bayes' theorem and estimate the marginal likelihood based on approximate representations of the joint model. Upon perfect approximation, the marginal likelihood is constant across all parameter values by definition. However, errors in approximate inference lead to undesirable variance in the marginal likelihood estimates across different parameter values. We penalize violations of this symmetry with a self-consistency loss which significantly improves the quality of approximate inference in low data regimes and can be used to augment the training of popular neural density estimators. We apply our method to a number of synthetic problems and realistic scientific models, discovering notable advantages in the context of both neural posterior and likelihood approximation.

We propose SelfVC, a training strategy to iteratively improve a voice conversion model with self-synthesized examples. Previous efforts on voice conversion focus on factorizing speech into explicitly disentangled representations that separately encode speaker characteristics and linguistic content. However, disentangling speech representations to capture such attributes using task-specific loss terms can lead to information loss. In this work, instead of explicitly disentangling attributes with loss terms, we present a framework to train a controllable voice conversion model on entangled speech representations derived from self-supervised learning (SSL) and speaker verification models. First, we develop techniques to derive prosodic information from the audio signal and SSL representations to train predictive submodules in the synthesis model. Next, we propose a training strategy to iteratively improve the synthesis model for voice conversion, by creating a challenging training objective using self-synthesized examples. We demonstrate that incorporating such self-synthesized examples during training improves the speaker similarity of generated speech as compared to a baseline voice conversion model trained solely on heuristically perturbed inputs. Our framework is trained without any text and achieves state-of-the-art results in zero-shot voice conversion on metrics evaluating naturalness, speaker similarity, and intelligibility of synthesized audio.

Raw images offer unique advantages in many low-level visual tasks due to their unprocessed nature. However, this unprocessed state accentuates noise, making raw images challenging to compress effectively. Current compression methods often overlook the ubiquitous noise in raw space, leading to increased bitrates and reduced quality. In this paper, we propose a novel raw image compression scheme that selectively compresses the noise-free component of the input, while discarding its real noise using a self-supervised approach. By excluding noise from the bitstream, both the coding efficiency and reconstruction quality are significantly enhanced. We curate an full-day dataset of raw images with calibrated noise parameters and reference images to evaluate the performance of models under a wide range of input signal-noise ratios. Experimental results demonstrate that our method surpasses existing compression techniques, achieving a more advantageous rate-distortion balance with improvements ranging from +2 to +10dB and yielding a bit saving of 2 to 50 times. The code will be released upon paper acceptance.

In this paper, we use matrix information theory as an analytical tool to analyze the dynamics of the information interplay between data representations and classification head vectors in the supervised learning process. Specifically, inspired by the theory of Neural Collapse, we introduce matrix mutual information ratio (MIR) and matrix entropy difference ratio (HDR) to assess the interactions of data representation and class classification heads in supervised learning, and we determine the theoretical optimal values for MIR and HDR when Neural Collapse happens. Our experiments show that MIR and HDR can effectively explain many phenomena occurring in neural networks, for example, the standard supervised training dynamics, linear mode connectivity, and the performance of label smoothing and pruning. Additionally, we use MIR and HDR to gain insights into the dynamics of grokking, which is an intriguing phenomenon observed in supervised training, where the model demonstrates generalization capabilities long after it has learned to fit the training data. Furthermore, we introduce MIR and HDR as loss terms in supervised and semi-supervised learning to optimize the information interactions among samples and classification heads. The empirical results provide evidence of the method's effectiveness, demonstrating that the utilization of MIR and HDR not only aids in comprehending the dynamics throughout the training process but can also enhances the training procedure itself.

Prior interpretability research studying narrow distributions has preliminarily identified self-repair, a phenomena where if components in large language models are ablated, later components will change their behavior to compensate. Our work builds off this past literature, demonstrating that self-repair exists on a variety of models families and sizes when ablating individual attention heads on the full training distribution. We further show that on the full training distribution self-repair is imperfect, as the original direct effect of the head is not fully restored, and noisy, since the degree of self-repair varies significantly across different prompts (sometimes overcorrecting beyond the original effect). We highlight two different mechanisms that contribute to self-repair, including changes in the final LayerNorm scaling factor and sparse sets of neurons implementing Anti-Erasure. We additionally discuss the implications of these results for interpretability practitioners and close with a more speculative discussion on the mystery of why self-repair occurs in these models at all, highlighting evidence for the Iterative Inference hypothesis in language models, a framework that predicts self-repair.

Learning in deep weight spaces (DWS), where neural networks process the weights of other neural networks, is an emerging research direction, with applications to 2D and 3D neural fields (INRs, NeRFs), as well as making inferences about other types of neural networks. Unfortunately, weight space models tend to suffer from substantial overfitting. We empirically analyze the reasons for this overfitting and find that a key reason is the lack of diversity in DWS datasets. While a given object can be represented by many different weight configurations, typical INR training sets fail to capture variability across INRs that represent the same object. To address this, we explore strategies for data augmentation in weight spaces and propose a MixUp method adapted for weight spaces. We demonstrate the effectiveness of these methods in two setups. In classification, they improve performance similarly to having up to 10 times more data. In self-supervised contrastive learning, they yield substantial 5-10% gains in downstream classification.

Subsurface imaging involves solving full waveform inversion (FWI) to predict geophysical properties from measurements. This problem can be reframed as an image-to-image translation, with the usual approach being to train an encoder-decoder network using paired data from two domains: geophysical property and measurement. A recent seminal work (InvLINT) demonstrates there is only a linear mapping between the latent spaces of the two domains, and the decoder requires paired data for training. This paper extends this direction by demonstrating that only linear mapping necessitates paired data, while both the encoder and decoder can be learned from their respective domains through self-supervised learning. This unveils an intriguing phenomenon (named Auto-Linear) where the self-learned features of two separate domains are automatically linearly correlated. Compared with existing methods, our Auto-Linear has four advantages: (a) solving both forward and inverse modeling simultaneously, (b) reducing model size, (c) enhanced performance, especially when the paired data is limited, and (d) strong generalization ability of the trained encoder and decoder.

Recent efforts have been made to integrate self-supervised learning (SSL) with the framework of federated learning (FL). One unique challenge of federated self-supervised learning (FedSSL) is that the global objective of FedSSL usually does not equal the weighted sum of local SSL objectives. Consequently, conventional approaches, such as federated averaging (FedAvg), fail to precisely minimize the FedSSL global objective, often resulting in suboptimal performance, especially when data is non-i.i.d.. To fill this gap, we propose a provable FedSSL algorithm, named FedSC, based on the spectral contrastive objective. In FedSC, clients share correlation matrices of data representations in addition to model weights periodically, which enables inter-client contrast of data samples in addition to intra-client contrast and contraction, resulting in improved quality of data representations. Differential privacy (DP) protection is deployed to control the additional privacy leakage on local datasets when correlation matrices are shared. We provide theoretical analysis on convergence and extra privacy leakage, and conduct numerical experiments to justify the effectiveness of our proposed algorithm.