Sign Operator for Coping with Heavy-Tailed Noise in Non-Convex Optimization: High Probability Bounds Under $(L_0, L_1)$-Smoothness

• URL: http://arxiv.org/abs/2502.07923v2
• Date: Tue, 27 May 2025 15:31:20 GMT
• Title: Sign Operator for Coping with Heavy-Tailed Noise in Non-Convex Optimization: High Probability Bounds Under $(L_0, L_1)$-Smoothness
• Authors: Nikita Kornilov, Philip Zmushko, Andrei Semenov, Mark Ikonnikov, Alexander Gasnikov, Alexander Beznosikov,
• Abstract summary: We show that SignSGD with Majority Voting can robustly work on the whole range of complexity with $kappakappakappakappa-1right, kappakappakappa-1right, kappakappakappa-1right, kappakappakappa-1right, kappakappakappa-1right, kappakappakappa-1right, kappakappakappa-1right, kappa
• Score: 74.18546828528298
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In recent years, non-convex optimization problems are more often described by generalized $(L_0, L_1)$-smoothness assumption rather than standard one. Meanwhile, severely corrupted data used in these problems has increased the demand for methods capable of handling heavy-tailed noises, i.e., noises with bounded $\kappa$-th moment. Motivated by these real-world trends and challenges, we explore sign-based methods in this setup and demonstrate their effectiveness in comparison with other popular solutions like clipping or normalization. In theory, we prove the first-known high probability convergence bounds under $(L_0, L_1)$-smoothness and heavy-tailed noises with mild parameter dependencies. In the case of standard smoothness, these bounds are novel for sign-based methods as well. In particular, SignSGD with batching achieves sample complexity $\tilde{O}\left(\left(\frac{\Delta L_0d}{\varepsilon^2} + \frac{\Delta L_1d^\frac{3}{2}}{\varepsilon}\right)\left[1 + \left(\frac{\sigma}{\varepsilon}\right)^\frac{\kappa}{\kappa-1}\right]\right), \kappa \in (1,2]$. Under the assumption of symmetric noises, SignSGD with Majority Voting can robustly work on the whole range of $\kappa \in (0,2]$ with complexity $\tilde{O}\left(\left(\frac{\Delta L_0d}{\varepsilon^2} + \frac{\Delta L_1d^\frac{3}{2}}{\varepsilon}\right)\left[\frac{1}{\kappa^2} + \frac{\sigma^2}{\varepsilon^2}\right]\right)$. We also obtain results for parameter-agnostic setups, Polyak-Lojasiewicz functions and momentum-based methods (in expectation). Our theoretical findings are supported by the superior performance of sign-based methods in training Large Language Models compared to clipping and normalization.

Related papers

• High Probability Complexity Bounds of Trust-Region Stochastic Sequential Quadratic Programming with Heavy-Tailed Noise [23.663813244183984]
  We propose a Trust-Region Sequential Quadratic Programming (TR-SSQP) method. We establish its high-probability complexity bounds for identifying first- and second-order $epsilon$-stationary points. We show that our method achieves the same high-probability first-order complexity bounds as in the light-tailed noise setting.
  arXiv  Detail & Related papers  (2025-03-24T19:23:13Z)
• Stochastic Halpern iteration in normed spaces and applications to reinforcement learning [0.30693357740321775]
  We show that if the underlying oracle is uniformly bounded, our method exhibits an overall oracle complexity of $tildeO(varepsilon-5)$. We propose new synchronous algorithms for average reward and discounted reward Markov decision processes.
  arXiv  Detail & Related papers  (2024-03-19T01:07:35Z)
• On Partial Optimal Transport: Revising the Infeasibility of Sinkhorn and Efficient Gradient Methods [17.14725907264431]
  This paper studies the Partial Optimal Transport (POT) problem between two unbalanced measures with at most $n$ supports. We propose a novel rounding algorithm for POT, and then provide a feasible Sinkhorn procedure with a revised complexity of $mathcalwidetilde O(n2/varepsilon4)$.
  arXiv  Detail & Related papers  (2023-12-21T15:56:09Z)
• Breaking the Heavy-Tailed Noise Barrier in Stochastic Optimization Problems [56.86067111855056]
  We consider clipped optimization problems with heavy-tailed noise with structured density. We show that it is possible to get faster rates of convergence than $mathcalO(K-(alpha - 1)/alpha)$, when the gradients have finite moments of order. We prove that the resulting estimates have negligible bias and controllable variance.
  arXiv  Detail & Related papers  (2023-11-07T17:39:17Z)
• A Specialized Semismooth Newton Method for Kernel-Based Optimal Transport [92.96250725599958]
  Kernel-based optimal transport (OT) estimators offer an alternative, functional estimation procedure to address OT problems from samples. We show that our SSN method achieves a global convergence rate of $O (1/sqrtk)$, and a local quadratic convergence rate under standard regularity conditions.
  arXiv  Detail & Related papers  (2023-10-21T18:48:45Z)
• Accelerating Inexact HyperGradient Descent for Bilevel Optimization [84.00488779515206]
  We present a method for solving general non-strongly-concave bilevel optimization problems. Our results also improve upon the existing complexity for finding second-order stationary points in non-strongly-concave problems.
  arXiv  Detail & Related papers  (2023-06-30T20:36:44Z)
• Beyond Uniform Smoothness: A Stopped Analysis of Adaptive SGD [38.221784575853796]
  This work considers the problem of finding first-order stationary point of a non atau function with potentially constant smoothness using a gradient. We develop a technique that allows us to prove $mathcalO(fracmathrmpolylog(T)sigmatT)$ convergence rates without assuming uniform bounds on the noise.
  arXiv  Detail & Related papers  (2023-02-13T18:13:36Z)
• Near-Optimal Non-Convex Stochastic Optimization under Generalized Smoothness [21.865728815935665]
  Two recent works established the $O(epsilon-3)$ sample complexity to obtain an $O(epsilon)$-stationary point. However, both require a large batch size on the order of $mathrmploy(epsilon-1)$, which is not only computationally burdensome but also unsuitable for streaming applications. In this work, we solve the prior two problems simultaneously by revisiting a simple variant of the STORM algorithm.
  arXiv  Detail & Related papers  (2023-02-13T00:22:28Z)
• A Fully First-Order Method for Stochastic Bilevel Optimization [8.663726907303303]
  We consider unconstrained bilevel optimization problems when only the first-order gradient oracles are available. We propose a Fully First-order Approximation (F2SA) method, and study its non-asymptotic convergence properties. We demonstrate even superior practical performance of the proposed method over existing second-order based approaches on MNIST data-hypercleaning experiments.
  arXiv  Detail & Related papers  (2023-01-26T05:34:21Z)
• An Efficient Stochastic Algorithm for Decentralized Nonconvex-Strongly-Concave Minimax Optimization [25.00475462213752]
  Decentralized Recursive desc. Method (DREAM) Concretely, it requires $mathcalO(minminappaappa3eps-3,kappa2 N)$ first-order oracle (SFO) calls and $tildemathcalO(kappa2 epsilon-2) communication rounds. Our numerical experiments validate the superiority of previous methods.
  arXiv  Detail & Related papers  (2022-12-05T16:09:39Z)
• Near Sample-Optimal Reduction-based Policy Learning for Average Reward MDP [58.13930707612128]
  This work considers the sample complexity of obtaining an $varepsilon$-optimal policy in an average reward Markov Decision Process (AMDP) We prove an upper bound of $widetilde O(H varepsilon-3 ln frac1delta)$ samples per state-action pair, where $H := sp(h*)$ is the span of bias of any optimal policy, $varepsilon$ is the accuracy and $delta$ is the failure probability.
  arXiv  Detail & Related papers  (2022-12-01T15:57:58Z)
• Tight Bounds on the Hardness of Learning Simple Nonparametric Mixtures [9.053430799456587]
  We study the problem of learning nonparametric distributions in a finite mixture. We establish tight bounds on the sample complexity for learning the component distributions in such models.
  arXiv  Detail & Related papers  (2022-03-28T23:53:48Z)
• Decentralized Stochastic Variance Reduced Extragradient Method [25.21457349137344]
  This paper studies decentralized convex-concave minimax optimization problems of the form $min_xmax_y fx,y triqfrac1msumi=1m f_i triqfrac1msumi=1m f_i triqfrac1msumi=1m f_i triqfrac1msumi=1m f_i triqfrac1msum
  arXiv  Detail & Related papers  (2022-02-01T16:06:20Z)
• Faster Rates of Differentially Private Stochastic Convex Optimization [7.93728520583825]
  We study the case where the population risk function satisfies the Tysbakov Noise Condition (TNC) with some parameter $theta>1$. In the second part, we focus on a special case where the population risk function is strongly convex.
  arXiv  Detail & Related papers  (2021-07-31T22:23:39Z)
• Hybrid Stochastic-Deterministic Minibatch Proximal Gradient: Less-Than-Single-Pass Optimization with Nearly Optimal Generalization [83.80460802169999]
  We show that HSDMPG can attain an $mathcalObig (1/sttnbig)$ which is at the order of excess error on a learning model. For loss factors, we prove that HSDMPG can attain an $mathcalObig (1/sttnbig)$ which is at the order of excess error on a learning model.
  arXiv  Detail & Related papers  (2020-09-18T02:18:44Z)
• Second-Order Information in Non-Convex Stochastic Optimization: Power and Limitations [54.42518331209581]
  We find an algorithm which finds. epsilon$-approximate stationary point (with $|nabla F(x)|le epsilon$) using. $(epsilon,gamma)$surimate random random points. Our lower bounds here are novel even in the noiseless case.
  arXiv  Detail & Related papers  (2020-06-24T04:41:43Z)
• Learning Halfspaces with Tsybakov Noise [50.659479930171585]
  We study the learnability of halfspaces in the presence of Tsybakov noise. We give an algorithm that achieves misclassification error $epsilon$ with respect to the true halfspace.
  arXiv  Detail & Related papers  (2020-06-11T14:25:02Z)
• Model-Free Reinforcement Learning: from Clipped Pseudo-Regret to Sample Complexity [59.34067736545355]
  Given an MDP with $S$ states, $A$ actions, the discount factor $gamma in (0,1)$, and an approximation threshold $epsilon > 0$, we provide a model-free algorithm to learn an $epsilon$-optimal policy. For small enough $epsilon$, we show an improved algorithm with sample complexity.
  arXiv  Detail & Related papers  (2020-06-06T13:34:41Z)
• Revisiting EXTRA for Smooth Distributed Optimization [70.65867695317633]
  We give a sharp complexity analysis for EXTRA with the improved $Oleft(left(fracLmu+frac11-sigma_2(W)right)logfrac1epsilon (1-sigma_2(W))right)$. Our communication complexities of the accelerated EXTRA are only worse by the factors of $left(logfracLmu (1-sigma_2(W))right)$ and $left(logfrac1epsilon (1-
  arXiv  Detail & Related papers  (2020-02-24T08:07:08Z)
• Curse of Dimensionality on Randomized Smoothing for Certifiable Robustness [151.67113334248464]
  We show that extending the smoothing technique to defend against other attack models can be challenging. We present experimental results on CIFAR to validate our theory.
  arXiv  Detail & Related papers  (2020-02-08T22:02:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.