AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Chaotic Regularization and Heavy-Tailed Limits for Deterministic Gradient Descent

Soon Hoe Lim, Yijun Wan, Umut Simsekli

NeurIPS 2022 · Optimization › Implicit bias and generalization · approach: SGD noise and flat minima · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use MPGD instead of full-batch GDResNet-18 · CIFAR-10 validationvalidation accuracy: 73.10 -> 75.97Improved
Use MPGD instead of full-batch GDResNet-18 · CIFAR-10accuracy gap: 8.99 -> 5.64Improved
Use MPGD instead of vanilla GDShallow neural net, width 16 · Airfoil Self-Noise testtest RMSE: 0.4309 -> 0.3810Improved
Use MPGD instead of vanilla GDShallow neural net, width 16 · Airfoil Self-NoiseRMSE gap: 0.2411 -> 0.2256Improved
Add uncorrelated Gaussian perturbations instead of MPGDFull-batch GD with ResNet-18 · CIFAR-10 validationvalidation accuracy: 73.10 -> 70.58Hurt
Set γ = 0.7 rather than γ = 0.6 (ablation)MPGD with shallow neural net · Airfoil Self-Noise testtest RMSE: 0.3810 -> 0.4600Hurt
Set γ = 0.55 rather than γ = 0.6 (ablation)MPGD with ResNet-18 · CIFAR-10 validationvalidation accuracy: 75.78 -> 75.97No clear effect
Increase γ from 0.6 to 0.7 (ablation)MPGD · Airfoil Self-Noise and CIFAR-10test performance: 0.3810 -> 0.4600; 75.78 -> 73.72Mixed
Replace MPGD perturbations with Gaussian perturbationsFull-batch GD · Widening valley lossoptimization successHurt
Use MPGD perturbations rather than Gaussian perturbationsFull-batch GD · CIFAR-10 validationvalidation accuracy: 70.58 -> 75.97Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy