Flavors of Margin: Implicit Bias of Steepest Descent in Homogeneous Neural Networks

We study the implicit bias of the general family of steepest descent algorithms, which includes gradient descent, sign descent and coordinate descent, in deep homogeneous neural networks. We prove that an algorithm-dependent geometric margin starts increasing once the networks reach perfect training accuracy and characterize the late-stage bias of the algorithms. In particular, we define a generalized notion of stationarity for optimization problems and show that the algorithms progressively reduce a (generalized) Bregman divergence, which quantifies proximity to such stationary points of a margin-maximization problem. We then experimentally zoom into the trajectories of neural networks optimized with various steepest descent algorithms, highlighting connections to the implicit bias of Adam.

On the Geometry of Regularization in Adversarial Training: High-Dimensional Asymptotics and Generalization Bounds

Regularization, whether explicit in terms of a penalty in the loss or implicit in the choice of algorithm, is a cornerstone of modern machine learning. Indeed, controlling the complexity of the model class is particularly important when data is scarce, noisy or contaminated, as it translates a statistical belief on the underlying structure of the data. This work investigates the question of how to choose the regularization norm $\lVert \cdot \rVert$ in the context of high-dimensional adversarial training for binary classification. To this end, we first derive an exact asymptotic description of the robust, regularized empirical risk minimizer for various types of adversarial attacks and regularization norms (including non-$\ell_p$ norms). We complement this analysis with a uniform convergence analysis, deriving bounds on the Rademacher Complexity for this class of problems. Leveraging our theoretical results, we quantitatively characterize the relationship between perturbation size and the optimal choice of $\lVert \cdot \rVert$, confirming the intuition that, in the data scarce regime, the type of regularization becomes increasingly important for adversarial training as perturbations grow in size.

Emergent properties with repeated examples

We study the performance of transformers as a function of the number of repetitions of training examples with algorithmically generated datasets. On three problems of mathematics: the greatest common divisor, modular multiplication, and matrix eigenvalues, we show that for a fixed number of training steps, models trained on smaller sets of repeated examples outperform models trained on larger sets of single-use examples. We also demonstrate that two-set training - repeated use of a small random subset of examples, along normal sampling on the rest of the training set - provides for faster learning and better performance. This highlights that the benefits of repetition can outweigh those of data diversity. These datasets and problems provide a controlled setting to shed light on the still poorly understood interplay between generalization and memorization in deep learning.

Strong Model Collapse

Within the scaling laws paradigm, which underpins the training of large neural networks like ChatGPT and Llama, we consider a supervised regression setting and establish the existance of a strong form of the model collapse phenomenon, a critical performance degradation due to synthetic data in the training corpus. Our results show that even the smallest fraction of synthetic data (e.g., as little as 1\% of the total training dataset) can still lead to model collapse: larger and larger training sets do not enhance performance. We further investigate whether increasing model size, an approach aligned with current trends in training large language models, exacerbates or mitigates model collapse. In a simplified regime where neural networks are approximated via random projections of tunable size, we both theoretically and empirically show that larger models can amplify model collapse. Interestingly, our theory also indicates that, beyond the interpolation threshold (which can be extremely high for very large datasets), larger models may mitigate the collapse, although they do not entirely prevent it. Our theoretical findings are empirically verified through experiments on language models and feed-forward neural networks for images.

Mission Impossible: A Statistical Perspective on Jailbreaking LLMs

Large language models (LLMs) are trained on a deluge of text data with limited quality control. As a result, LLMs can exhibit unintended or even harmful behaviours, such as leaking information, fake news or hate speech. Countermeasures, commonly referred to as preference alignment, include fine-tuning the pretrained LLMs with carefully crafted text examples of desired behaviour. Even then, empirical evidence shows preference aligned LLMs can be enticed to harmful behaviour. This so called jailbreaking of LLMs is typically achieved by adversarially modifying the input prompt to the LLM. Our paper provides theoretical insights into the phenomenon of preference alignment and jailbreaking from a statistical perspective. Under our framework, we first show that pretrained LLMs will mimic harmful behaviour if present in the training corpus. Under that same framework, we then introduce a statistical notion of alignment, and lower-bound the jailbreaking probability, showing that it is unpreventable under reasonable assumptions. Based on our insights, we propose an alteration to the currently prevalent alignment strategy RLHF. Specifically, we introduce a simple modification to the RLHF objective, we call E-RLHF, that aims to increase the likelihood of safe responses. E-RLHF brings no additional training cost, and is compatible with other methods. Empirically, we demonstrate that E-RLHF outperforms RLHF on all alignment problems put forward by the AdvBench and HarmBench project without sacrificing model performance as measured by the MT-Bench project.

Beyond Model Collapse: Scaling Up with Synthesized Data Requires Reinforcement

Synthesized data from generative models is increasingly considered as an alternative to human-annotated data for fine-tuning Large Language Models. This raises concerns about model collapse: a drop in performance of models fine-tuned on generated data. Considering that it is easier for both humans and machines to tell between good and bad examples than to generate high-quality samples, we investigate the use of feedback on synthesized data to prevent model collapse. We derive theoretical conditions under which a Gaussian mixture classification model can achieve asymptotically optimal performance when trained on feedback-augmented synthesized data, and provide supporting simulations for finite regimes. We illustrate our theoretical predictions on two practical problems: computing matrix eigenvalues with transformers and news summarization with large language models, which both undergo model collapse when trained on model-generated data. We show that training from feedback-augmented synthesized data, either by pruning incorrect predictions or by selecting the best of several guesses, can prevent model collapse, validating popular approaches like RLHF.

The Price of Implicit Bias in Adversarially Robust Generalization

We study the implicit bias of optimization in robust empirical risk minimization (robust ERM) and its connection with robust generalization. In classification settings under adversarial perturbations with linear models, we study what type of regularization should ideally be applied for a given perturbation set to improve (robust) generalization. We then show that the implicit bias of optimization in robust ERM can significantly affect the robustness of the model and identify two ways this can happen; either through the optimization algorithm or the architecture. We verify our predictions in simulations with synthetic data and experimentally study the importance of implicit bias in robust ERM with deep neural networks.

Iteration Head: A Mechanistic Study of Chain-of-Thought

Chain-of-Thought (CoT) reasoning is known to improve Large Language Models both empirically and in terms of theoretical approximation power. However, our understanding of the inner workings and conditions of apparition of CoT capabilities remains limited. This paper helps fill this gap by demonstrating how CoT reasoning emerges in transformers in a controlled and interpretable setting. In particular, we observe the appearance of a specialized attention mechanism dedicated to iterative reasoning, which we coined "iteration heads". We track both the emergence and the precise working of these iteration heads down to the attention level, and measure the transferability of the CoT skills to which they give rise between tasks.

Attacking Bayes: On the Adversarial Robustness of Bayesian Neural Networks

Adversarial examples have been shown to cause neural networks to fail on a wide range of vision and language tasks, but recent work has claimed that Bayesian neural networks (BNNs) are inherently robust to adversarial perturbations. In this work, we examine this claim. To study the adversarial robustness of BNNs, we investigate whether it is possible to successfully break state-of-the-art BNN inference methods and prediction pipelines using even relatively unsophisticated attacks for three tasks: (1) label prediction under the posterior predictive mean, (2) adversarial example detection with Bayesian predictive uncertainty, and (3) semantic shift detection. We find that BNNs trained with state-of-the-art approximate inference methods, and even BNNs trained with Hamiltonian Monte Carlo, are highly susceptible to adversarial attacks. We also identify various conceptual and experimental errors in previous works that claimed inherent adversarial robustness of BNNs and conclusively demonstrate that BNNs and uncertainty-aware Bayesian prediction pipelines are not inherently robust against adversarial attacks.

Robust Data Pruning: Uncovering and Overcoming Implicit Bias

In the era of exceptionally data-hungry models, careful selection of the training data is essential to mitigate the extensive costs of deep learning. Data pruning offers a solution by removing redundant or uninformative samples from the dataset, which yields faster convergence and improved neural scaling laws. However, little is known about its impact on classification bias of the trained models. We conduct the first systematic study of this effect and reveal that existing data pruning algorithms can produce highly biased classifiers. At the same time, we argue that random data pruning with appropriate class ratios has potential to improve the worst-class performance. We propose a "fairness-aware" approach to pruning and empirically demonstrate its performance on standard computer vision benchmarks. In sharp contrast to existing algorithms, our proposed method continues improving robustness at a tolerable drop of average performance as we prune more from the datasets. We present theoretical analysis of the classification risk in a mixture of Gaussians to further motivate our algorithm and support our findings.