Practical Sharpness-Aware Minimization Cannot Converge All the Way to Optima

• URL: http://arxiv.org/abs/2306.09850v3
• Date: Fri, 27 Oct 2023 05:08:13 GMT
• Title: Practical Sharpness-Aware Minimization Cannot Converge All the Way to Optima
• Authors: Dongkuk Si, Chulhee Yun
• Abstract summary: Sharpness-Aware Minimization (SAM) is an that takes a step based on a perturbation at a $y_t = x_t + rho fracbla f(x_t)lt blablax_t)
• Score: 14.141453107129403
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Sharpness-Aware Minimization (SAM) is an optimizer that takes a descent step based on the gradient at a perturbation $y_t = x_t + \rho \frac{\nabla f(x_t)}{\lVert \nabla f(x_t) \rVert}$ of the current point $x_t$. Existing studies prove convergence of SAM for smooth functions, but they do so by assuming decaying perturbation size $\rho$ and/or no gradient normalization in $y_t$, which is detached from practice. To address this gap, we study deterministic/stochastic versions of SAM with practical configurations (i.e., constant $\rho$ and gradient normalization in $y_t$) and explore their convergence properties on smooth functions with (non)convexity assumptions. Perhaps surprisingly, in many scenarios, we find out that SAM has limited capability to converge to global minima or stationary points. For smooth strongly convex functions, we show that while deterministic SAM enjoys tight global convergence rates of $\tilde \Theta(\frac{1}{T^2})$, the convergence bound of stochastic SAM suffers an inevitable additive term $O(\rho^2)$, indicating convergence only up to neighborhoods of optima. In fact, such $O(\rho^2)$ factors arise for stochastic SAM in all the settings we consider, and also for deterministic SAM in nonconvex cases; importantly, we prove by examples that such terms are unavoidable. Our results highlight vastly different characteristics of SAM with vs. without decaying perturbation size or gradient normalization, and suggest that the intuitions gained from one version may not apply to the other.

Related papers

• Transformer Injectivity & Geometric Robustness - Analytic Margins and Bi-Lipschitz Uniformity of Sequence-Level Hidden States [0.0]
  We show that the map from discrete prompts to last-token hidden states is generically injective on finite prompt sets.<n>We study behavior across layers, sequence lengths, model scales, and 8- and 4-bit activation quantization.
  arXiv  Detail & Related papers  (2025-11-17T19:39:15Z)
• Beyond likelihood ratio bias: Nested multi-time-scale stochastic approximation for likelihood-free parameter estimation [49.78792404811239]
  We study inference in simulation-based models where the analytical form of the likelihood is unknown.<n>We use a ratio-free nested multi-time-scale approximation (SA) method that simultaneously tracks the score and drives the parameter update.<n>We show that our algorithm can eliminate the original bias $Obig(sqrtfrac1Nbig)$ and accelerate the convergence rate from $Obig(beta_k+sqrtfracalpha_kNbig)$.
  arXiv  Detail & Related papers  (2024-11-20T02:46:15Z)
• Bilateral Sharpness-Aware Minimization for Flatter Minima [61.17349662062522]
  Sharpness-Aware Minimization (SAM) enhances generalization by reducing a Max-Sharpness (MaxS) In this paper, we propose to utilize the difference between the training loss and the minimum loss over the neighborhood surrounding the current weight, which we denote as Min-Sharpness (MinS) By merging MaxS and MinS, we created a better FI that indicates a flatter direction during the optimization. Specially, we combine this FI with SAM into the proposed Bilateral SAM (BSAM) which finds a more flatter minimum than that of SAM.
  arXiv  Detail & Related papers  (2024-09-20T03:01:13Z)
• Transfer Operators from Batches of Unpaired Points via Entropic Transport Kernels [3.099885205621181]
  We derive a maximum-likelihood inference functional, propose a computationally tractable approximation and analyze their properties. We prove a $Gamma$-convergence result showing that we can recover the true density from empirical approximations as the number $N$ of blocks goes to infinity.
  arXiv  Detail & Related papers  (2024-02-13T12:52:41Z)
• Convergence Rates for Stochastic Approximation: Biased Noise with Unbounded Variance, and Applications [2.0584253077707477]
  We study the convergence properties of the Gradient Descent (SGD) method for finding a stationary point of an objective function $J(cdot)$. Our results apply to a class of invex'' functions, which have the property that every stationary point is also a global minimizer.
  arXiv  Detail & Related papers  (2023-12-05T15:22:39Z)
• An Oblivious Stochastic Composite Optimization Algorithm for Eigenvalue Optimization Problems [76.2042837251496]
  We introduce two oblivious mirror descent algorithms based on a complementary composite setting. Remarkably, both algorithms work without prior knowledge of the Lipschitz constant or smoothness of the objective function. We show how to extend our framework to scale and demonstrate the efficiency and robustness of our methods on large scale semidefinite programs.
  arXiv  Detail & Related papers  (2023-06-30T08:34:29Z)
• Achieving the Asymptotically Optimal Sample Complexity of Offline Reinforcement Learning: A DRO-Based Approach [36.88301225561535]
  offline reinforcement learning aims to learn from pre-collected datasets without active exploration. Existing approaches adopt a pessimistic stance towards uncertainty by penalizing rewards of under-explored state-action pairs to estimate value functions conservatively. We show that the distributionally robust optimization (DRO) based approach can also address these challenges and is asymptotically minimax optimal
  arXiv  Detail & Related papers  (2023-05-22T17:50:18Z)
• Best Policy Identification in Linear MDPs [70.57916977441262]
  We investigate the problem of best identification in discounted linear Markov+Delta Decision in the fixed confidence setting under a generative model. The lower bound as the solution of an intricate non- optimization program can be used as the starting point to devise such algorithms.
  arXiv  Detail & Related papers  (2022-08-11T04:12:50Z)
• Randomized Coordinate Subgradient Method for Nonsmooth Composite Optimization [11.017632675093628]
  Coordinate-type subgradient methods for addressing nonsmooth problems are relatively underexplored due to the set of properties of the Lipschitz-type assumption.
  arXiv  Detail & Related papers  (2022-06-30T02:17:11Z)
• Sharper Rates and Flexible Framework for Nonconvex SGD with Client and Data Sampling [64.31011847952006]
  We revisit the problem of finding an approximately stationary point of the average of $n$ smooth and possibly non-color functions. We generalize the $smallsfcolorgreen$ so that it can provably work with virtually any sampling mechanism. We provide the most general and most accurate analysis of optimal bound in the smooth non-color regime.
  arXiv  Detail & Related papers  (2022-06-05T21:32:33Z)
• Thinking Outside the Ball: Optimal Learning with Gradient Descent for Generalized Linear Stochastic Convex Optimization [37.177329562964765]
  We consider linear prediction with a convex Lipschitz loss, or more generally, convex optimization problems of generalized linear form. We show that in this setting, early iteration stopped Gradient Descent (GD), without any explicit regularization or projection, ensures excess error at most $epsilon$. But instead of uniform convergence in a norm ball, which we show can guarantee suboptimal learning using $Theta (1/epsilon4)$ samples, we rely on uniform convergence in a distribution-dependent ball.
  arXiv  Detail & Related papers  (2022-02-27T09:41:43Z)
• Sample Complexity of Asynchronous Q-Learning: Sharper Analysis and Variance Reduction [63.41789556777387]
  Asynchronous Q-learning aims to learn the optimal action-value function (or Q-function) of a Markov decision process (MDP) We show that the number of samples needed to yield an entrywise $varepsilon$-accurate estimate of the Q-function is at most on the order of $frac1mu_min (1-gamma)5varepsilon2+ fract_mixmu_min (1-gamma)$ up to some logarithmic factor.
  arXiv  Detail & Related papers  (2020-06-04T17:51:00Z)
• A Simple Convergence Proof of Adam and Adagrad [74.24716715922759]
  We show a proof of convergence between the Adam Adagrad and $O(d(N)/st)$ algorithms. Adam converges with the same convergence $O(d(N)/st)$ when used with the default parameters.
  arXiv  Detail & Related papers  (2020-03-05T01:56:17Z)

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.