---
title: "Sharpness-Aware Minimization for Efficiently Improving Generalization"
description: "Sharpness-Aware Minimization trains for a low-loss neighborhood rather than one low-loss point. Follow the two-pass update, the rho-sized adversarial weight probe, compute-aware vision results, official implementation, and the later theory that narrows the flatness story."
canonical_url: "https://fanout.sh/daily/2026-11-19-sharpness-aware-minimization"
md_url: "https://fanout.sh/daily/2026-11-19-sharpness-aware-minimization.md"
last_updated: "2026-11-19"
access: "public"
---

# Sharpness-Aware Minimization for Efficiently Improving Generalization

Sharpness-Aware Minimization trains for a low-loss neighborhood rather than one low-loss point. Follow the two-pass update, the rho-sized adversarial weight probe, compute-aware vision results, official implementation, and the later theory that narrows the flatness story.

## Paper details

- Authors: Pierre Foret, Ariel Kleiner, Hossein Mobahi, and Behnam Neyshabur

- Venue: ICLR 2021

- Track: AI Research

- Difficulty: Medium

- Reading time: 10 min

- Original paper: https://arxiv.org/abs/2010.01412v3

## What you will learn

- How SAM turns a worst-case loss inside a small weight-space neighborhood into a practical two-gradient update.

- Why the first pass only locates a high-loss perturbation while the second gradient updates the original weights.

- What the reported CIFAR, ImageNet, finetuning, and noisy-label gains establish, and where extra compute, rho tuning, parameterization, and later theory limit the claim.

## Continue exploring

[Revisit deep learning generalization](https://fanout.sh/daily/2026-10-09-rethinking-generalization): Connect SAM's neighborhood objective to the experiment showing that identical training fit can coexist with radically different held-out behavior.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md
