analiza aiqbench
This research addresses the challenge of training large models under heavy-tailed noise, a common issue in reinforcement learning and LLM development. The authors propose a new nested-loop algorithm that achieves theoretical complexity bounds matching existing methods while maintaining empirical superiority. For practitioners dealing with unstable gradient estimates, this work offers a robust optimization framework specifically tailored for non-convex bilevel problems.
Stochastic Bilevel Optimization with Heavy-Tailed Noise
arXiv:2509.14952v3 Announce Type: replace Abstract: This paper considers the smooth bilevel optimization in which the lower-level problem is strongly convex and the upper-level problem is possibly nonconvex. We focus on the stochastic setting where the algorithm can access the unbiased stochastic gradient evaluation with heavy-tailed noise, which is prevalent in many machine learning applications, such as training large language models and reinforcement learning. We propose a nested-loop normalized stochastic bilevel approximation (N$^2$SBA) for finding an $\epsilon$-stationary point with the stochastic first-order oracle (SFO) complexity of $\tilde{\mathcal{O}}\big(\kappa^{\frac{7p-3}{p-1}} \sigma^{\frac{p}{p-1}} \epsilon^{-\frac{4 p - 2}{p-1}}\big)$, where $\kappa$ is the condition number, $p\in(1,2]$ is the order of central moment for the noise, and $\sigma$ is the noise level. Furthermore, we specialize our idea to solve the nonconvex-strongly-concave minimax optimization problem, achieving an $\epsilon$-stationary point with the SFO complexity of~$\tilde{\mathcal O}\big(\kappa^{\frac{2p-1}{p-1}} \sigma^{\frac{p}{p-1}} \epsilon^{-\frac{3p-2}{p-1}}\big)$. All the above upper bounds match the best-known results under the special case of the bounded variance setting, i.e., $p=2$. We also conduct the numerical experiments to show the empirical superiority of the proposed methods.
Abstract:This paper considers the smooth bilevel optimization in which the lower-level problem is strongly convex and the upper-level problem is possibly nonconvex. We focus on the stochastic setting where the algorithm can access the unbiased stochastic gradient evaluation with heavy-tailed noise, which is prevalent in many machine learning applications, such as training large language models and reinforcement learning. We propose a nested-loop normalized stochastic bilevel approximation (N$^2$SBA) for finding an $\epsilon$-stationary point with the stochastic first-order oracle (SFO) complexity of $\tilde{\mathcal{O}}\big(\kappa^{\frac{7p-3}{p-1}} \sigma^{\frac{p}{p-1}} \epsilon^{-\frac{4 p - 2}{p-1}}\big)$, where $\kappa$ is the condition number, $p\in(1,2]$ is the order of central moment for the noise, and $\sigma$ is the noise level. Furthermore, we specialize our idea to solve the nonconvex-strongly-concave minimax optimization problem, achieving an $\epsilon$-stationary point with the SFO complexity of~$\tilde{\mathcal O}\big(\kappa^{\frac{2p-1}{p-1}} \sigma^{\frac{p}{p-1}} \epsilon^{-\frac{3p-2}{p-1}}\big)$. All the above upper bounds match the best-known results under the special case of the bounded variance setting, i.e., $p=2$. We also conduct the numerical experiments to show the empirical superiority of the proposed methods.
Machine Learning (cs.LG)
arXiv:2509.14952 [cs.LG]
(or arXiv:2509.14952v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2509.14952 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 259 cuvinte
Subiecte conexe
Accelerated stochastic first-order method for convex optimization under heavy-tailed noise
arXiv · 22 sept. 2026
The Role of Gradient Modification in Heavy-Tailed Nonconvex Stochastic Min-Max Optimization
arXiv · 9 sept. 2026
Stability and Generalization of Stochastic Nonconvex Optimization under Heavy-Tailed Noise: From Gradient Norm to Goldstein Stationarity
arXiv · 6 oct. 2026
Tight Complexity of Muon under Heavy-Tailed Noise
arXiv · 14 oct. 2026