Johannes Maly

Christian Kümmerle, Johannes Maly

We propose a new algorithm for the problem of recovering data that adheres to multiple, heterogeneous low-dimensional structures from linear observations. Focusing on data matrices that are simultaneously row-sparse and low-rank, we propose and analyze an iteratively reweighted least squares (IRLS) algorithm that is able to leverage both structures. In particular, it optimizes a combination of non-convex surrogates for row-sparsity and rank, a balancing of which is built into the algorithm. We prove locally quadratic convergence of the iterates to a simultaneously structured data matrix in a regime of minimal sample complexity (up to constants and a logarithmic factor), which is known to be impossible for a combination of convex surrogates. In experiments, we show that the IRLS method exhibits favorable empirical convergence, identifying simultaneously row-sparse and low-rank matrices from fewer measurements than state-of-the-art methods. Code is available at https://github.com/ckuemmerle/simirls.

Johannes Maly, Rayan Saab

In this short note, we propose a new method for quantizing the weights of a fully trained neural network. A simple deterministic pre-processing step allows us to quantize network layers via memoryless scalar quantization while preserving the network performance on given training data. On one hand, the computational complexity of this pre-processing slightly exceeds that of state-of-the-art algorithms in the literature. On the other hand, our approach does not require any hyper-parameter tuning and, in contrast to previous methods, allows a plain analysis. We provide rigorous theoretical guarantees in the case of quantizing single network layers and show that the relative error decays with the number of parameters in the network if the training data behaves well, e.g., if it is sampled from suitable random distributions. The developed method also readily allows the quantization of deep networks by consecutive application to single layers.

Johannes Maly, Anna Veselovska

Graph data are ubiquitous in natural sciences and machine learning. In this paper, we consider the problem of quantizing graph structured, bandlimited data to few bits per entry while preserving its information under low-pass filtering. We propose an efficient single-shot noise shaping method that achieves state-of-the-art performance and comes with rigorous error bounds. In contrast to existing methods it allows reliable quantization to arbitrary bit-levels including the extreme case of using a single bit per data coefficient.

Vit Fojtik, Maria Matveev, Hung-Hsu Chou et al.

A widely believed explanation for the remarkable generalization capacities of overparameterized neural networks is that the optimization algorithms used for training induce an implicit bias towards benign solutions. To grasp this theoretically, recent works examine gradient descent and its variants in simplified training settings, often assuming vanishing learning rates. These studies reveal various forms of implicit regularization, such as $\ell_1$-norm minimizing parameters in regression and max-margin solutions in classification. Concurrently, empirical findings show that moderate to large learning rates exceeding standard stability thresholds lead to faster, albeit oscillatory, convergence in the so-called Edge-of-Stability regime, and induce an implicit bias towards minima of low sharpness (norm of training loss Hessian). In this work, we argue that a comprehensive understanding of the generalization performance of gradient descent requires analyzing the interaction between these various forms of implicit regularization. We empirically demonstrate that the learning rate balances between low parameter norm and low sharpness of the trained model. We furthermore prove for diagonal linear networks trained on a simple regression task that neither implicit bias alone minimizes the generalization error. These findings demonstrate that focusing on a single implicit bias is insufficient to explain good generalization, and they motivate a broader view of implicit regularization that captures the dynamic trade-off between norm and sharpness induced by non-negligible learning rates.

Hung-Hsu Chou, Johannes Maly, Holger Rauhut

In deep learning it is common to overparameterize neural networks, that is, to use more parameters than training samples. Quite surprisingly training the neural network via (stochastic) gradient descent leads to models that generalize very well, while classical statistics would suggest overfitting. In order to gain understanding of this implicit bias phenomenon we study the special case of sparse recovery (compressed sensing) which is of interest on its own. More precisely, in order to reconstruct a vector from underdetermined linear measurements, we introduce a corresponding overparameterized square loss functional, where the vector to be reconstructed is deeply factorized into several vectors. We show that, if there exists an exact solution, vanilla gradient flow for the overparameterized loss functional converges to a good approximation of the solution of minimal $\ell_1$-norm. The latter is well-known to promote sparse solutions. As a by-product, our results significantly improve the sample complexity for compressed sensing via gradient flow/descent on overparameterized models derived in previous works. The theory accurately predicts the recovery rate in numerical experiments. Our proof relies on analyzing a certain Bregman divergence of the flow. This bypasses the obstacles caused by non-convexity and should be of independent interest.

Hung-Hsu Chou, Carsten Gieshoff, Johannes Maly et al.

In deep learning, it is common to use more network parameters than training points. In such scenarioof over-parameterization, there are usually multiple networks that achieve zero training error so that thetraining algorithm induces an implicit bias on the computed solution. In practice, (stochastic) gradientdescent tends to prefer solutions which generalize well, which provides a possible explanation of thesuccess of deep learning. In this paper we analyze the dynamics of gradient descent in the simplifiedsetting of linear networks and of an estimation problem. Although we are not in an overparameterizedscenario, our analysis nevertheless provides insights into the phenomenon of implicit bias. In fact, wederive a rigorous analysis of the dynamics of vanilla gradient descent, and characterize the dynamicalconvergence of the spectrum. We are able to accurately locate time intervals where the effective rankof the iterates is close to the effective rank of a low-rank projection of the ground-truth matrix. Inpractice, those intervals can be used as criteria for early stopping if a certain regularity is desired. Wealso provide empirical evidence for implicit bias in more general scenarios, such as matrix sensing andrandom initialization. This suggests that deep learning prefers trajectories whose complexity (measuredin terms of effective rank) is monotonically increasing, which we believe is a fundamental concept for thetheoretical understanding of deep learning.