AI Paper Atlas

Research map / Learning theory and algorithms / Matrix and tensor methods

EA-CG: An Approximate Second-Order Method for Training Fully-Connected Neural Networks

Sheng-Wei Chen, Chun-Nan Chou, Edward Y. Chang

AAAI 2019 · Learning theory and algorithms › Matrix and tensor methods · approach: Geometry of low-rank and SPD structure · 6 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 PCH curvature with EA-CG instead of KFIFCNN training · Cifar-10, cross-entropytesting accuracyImproved
Use EA-CG instead of KFIFCNN training · Cifar-10, cross-entropywall-clock convergenceImproved
Use EA-CG instead of KFIFCNN training · ImageNet-10, cross-entropywall-clock timeNo clear effect
Use PCH-1 or PCH-2 rather than Fisher or Gauss-Newton curvatureFCNN Hessian approximation · Cifar-10, cross-entropytotal Hessian error: 0.1535 vs 0.0402; 0.0446 vs 0.0330Improved
Use PCH-1 or PCH-2 rather than Fisher or Gauss-Newton curvatureFCNN Hessian approximation · ImageNet-10, cross-entropytotal Hessian error: 0.0910 vs 0.0424; 0.0436 vs 0.0369Improved
Use PCH-1 rather than Fisher curvatureFCNN Hessian approximation · Cifar-10, non-convex criteriontotal Hessian error: 0.1198 vs 0.0349Improved
Use PCH-1 rather than Fisher curvatureFCNN Hessian approximation · ImageNet-10, non-convex criteriontotal Hessian error: 0.0880 vs 0.0598Improved
Use the PCH-1 eigenvalue treatment rather than PCH-2 (ablation)FCNN Hessian approximation · Cifar-10, cross-entropytotal Hessian error: 0.0402 vs 0.0330Hurt
Use the PCH-1 eigenvalue treatment rather than PCH-2 (ablation)FCNN Hessian approximation · ImageNet-10, cross-entropytotal Hessian error: 0.0424 vs 0.0369Hurt
Use the PCH-1 eigenvalue treatment rather than PCH-2 (ablation)FCNN Hessian approximation · Cifar-10, non-convex criteriontotal Hessian error: 0.0853 vs 0.0349Improved
Use the PCH-1 eigenvalue treatment rather than PCH-2 (ablation)FCNN Hessian approximation · ImageNet-10, non-convex criteriontotal Hessian error: 0.0660 vs 0.0598Improved
Use PCH-1 instead of PCH-2 (ablation)FCNN training · Cifar-10 and ImageNet-10, convex and non-convex criteriatraining loss and testing accuracyNo clear effect

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