paper-with-me

홈 › Papers

Generalization Bounds for Heavy-Tailed SDEs through the Fractional Fokker-Planck Equation

2024-02-12 · Benjamin Dupuis, Umut Şimşekli

Understanding the generalization properties of heavy-tailed stochastic optimization algorithms has attracted increasing attention over the past years. While illuminating interesting aspects of stochastic optimizers by using heavy-tailed stochastic differential equations as proxies, prior works either provided expected generalization bounds, or introduced non-computable information theoretic terms. Addressing these drawbacks, in this work, we prove high-probability generalization bounds for heavy-tailed SDEs which do not contain any nontrivial information theoretic terms. To achieve this goal, we develop new proof techniques based on estimating the entropy flows associated with the so-called fractional Fokker-Planck equation (a partial differential equation that governs the evolution of the distribution of the corresponding heavy-tailed SDE). In addition to obtaining high-probability bounds, we show that our bounds have a better dependence on the dimension of parameters as compared to prior art. Our results further identify a phase transition phenomenon, which suggests that heavy tails can be either beneficial or harmful depending on the problem structure. We support our theory with experiments conducted in a variety of settings.

📄 PDF Abstract BibTeX arXiv:2402.07723

Code (1)

benjidupuis/heavy_tails_generalization 공식 구현 pytorch

Tasks

Generalization BoundsStochastic Optimization

Similar Papers 제목 키워드 기반

Algorithmic Stability of Heavy-Tailed SGD with General Loss Functions

2023-01-27 · Anant Raj, Lingjiong Zhu, Mert Gürbüzbalaban, Umut Şimşekli

Heavy-tail phenomena in stochastic gradient descent (SGD) have been reported in several empirical studies. Experimental evidence in previous works suggests a strong interplay between the heaviness of the tails and genera…

Generalization Bounds

Algorithmic Stability of Stochastic Gradient Descent with Momentum under Heavy-Tailed Noise

2025-02-02 · Thanh Dang, Melih Barsbey, A K M Rokonuzzaman Sonet, Mert Gurbuzbalaban 외

Understanding the generalization properties of optimization algorithms under heavy-tailed noise has gained growing attention. However, the existing theoretical results mainly focus on stochastic gradient descent (SGD) an…

Generalization Bounds

Rényi Differential Privacy for Heavy-Tailed SDEs via Fractional Poincaré Inequalities

2025-11-19 · Benjamin Dupuis, Mert Gürbüzbalaban, Umut Şimşekli, Jian Wang 외 arxiv

Characterizing the differential privacy (DP) of learning algorithms has become a major challenge in recent years. In parallel, many studies suggested investigating the behavior of stochastic gradient descent (SGD) with h…

Hausdorff Dimension, Heavy Tails, and Generalization in Neural Networks

2020-06-16 · NeurIPS 2020 12 · Umut Şimşekli, Ozan Sener, George Deligiannidis, Murat A. Erdogdu

Despite its success in a wide range of applications, characterizing the generalization properties of stochastic gradient descent (SGD) in non-convex deep learning problems is still an important challenge. While modeling …

Generalization Bounds

Stability and Generalization of Nonconvex Optimization with Heavy-Tailed Noise

2026-01-27 · Hongxu Chen, Ke Wei, Xiaoming Yuan, Luo Luo arxiv

The empirical evidence indicates that stochastic optimization with heavy-tailed gradient noise is more appropriate to characterize the training of machine learning models than that with standard bounded gradient variance…

Stochastic Optimization