Research map / Optimization / Implicit bias and generalization
Entropy-SGD: Biasing Gradient Descent Into Wide Valleys
Pratik Chaudhari, Anna Choromanska, Stefano Soatto, Yann LeCun, Carlo Baldassi, Christian Borgs, Jennifer Chayes, Levent Sagun, Riccardo Zecchina
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Optimize local entropy with Entropy-SGD rather than the standard objective | mnistfc on MNIST · MNIST | test error (%): 1.39 ± 0.03 -> 1.37 ± 0.03 | No clear effect |
| Optimize local entropy with Entropy-SGD rather than the standard objective | LeNet · MNIST | test error (%): 0.51 ± 0.01 -> 0.5 ± 0.01 | No clear effect |
| Optimize local entropy with Entropy-SGD rather than the standard objective | All-CNN-BN · CIFAR-10 | test error (%): 7.71 ± 0.19 -> 7.81 ± 0.09 | No clear effect |
| Optimize local entropy with Entropy-SGD rather than the standard objective | PTB-LSTM · PTB test | perplexity: 78.6 ± 0.26 -> 77.656 ± 0.171 | Improved |
| Optimize local entropy with Entropy-Adam rather than Adam | char-LSTM · War and Peace test | perplexity: 1.226 ± 0.01 -> 1.217 ± 0.005 | Improved |
| Use Entropy-SGD instead of SGD | PTB-LSTM · PTB | epochs: 55 -> 25 | Improved |
| Use Entropy-Adam instead of Adam | char-LSTM · War and Peace | epochs: 40 -> 25 | Improved |
| Replace the local-entropy objective with vanilla SGLD | LeNet · MNIST test | test error (%): 0.50 ± 0.01 -> 0.63 ± 0.1 | Hurt |
| Replace the local-entropy objective with vanilla SGLD | All-CNN-BN · CIFAR-10 test | test error (%): 7.81 ± 0.09 -> 9.89 ± 0.11 | Hurt |
| Replace the local-entropy objective with vanilla SGLD | PTB-LSTM · PTB test | perplexity: 77.656 ± 0.171 -> 94.03 | Hurt |
| Replace the local-entropy objective with vanilla SGLD | char-LSTM · War and Peace test | perplexity: 1.213 ± 0.007 -> 1.3375 | Hurt |
| Anneal the SGLD learning rate (including matching it to the outer learning rate) (ablation) | Entropy-SGD · Deep-network experiments | local-entropy gradient estimate quality | Hurt |
Later papers that compared with this method
1 later accepted paper reports Entropy-SGD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Flatness-Aware Stochastic Gradient Langevin Dynamics (ICML 2026 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Entropic gradient descent algorithms and wide flat minima (ICLR 2021 · similarity 0.91)
- SGD Can Converge to Local Maxima (ICLR 2022 · similarity 0.89)
- A Unified Stability Analysis of SAM vs SGD: Role of Data Coherence and Emergence of Simplicity Bias (NeurIPS 2025 · similarity 0.88)
- A Precise Characterization of SGD Stability Using Loss Surface Geometry (ICLR 2024 · similarity 0.88)
- The Global Convergence Time of Stochastic Gradient Descent in Non-Convex Landscapes: Sharp Estimates via Large Deviations (ICML 2025 · similarity 0.87)
- Emergence and scaling laws in SGD learning of shallow neural networks (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
