AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Manifold Neural Network with Non-Gradient Optimization

Rui Zhang, Ziheng Jiao, Hongyuan Zhang, Xuelong Li

TPAMI 2022 · Optimization › Implicit bias and generalization · approach: Neural tangent kernel analysis · 9 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
replace gradient-trained DNN with ridge-regression reconstructed network (ablation)DNN with softmax · AT&Taccuracy: 50.89 -> 90.86Improved
replace gradient-trained DNN with ridge-regression reconstructed network (ablation)DNN with softmax · UMISTaccuracy: 72.83 -> 93.91Improved
replace gradient-trained DNN with ridge-regression reconstructed network (ablation)DNN with softmax · WAVEFORMaccuracy: 83.98 -> 68.49Hurt
add flexible Stiefel manifold SVM decision layer (ablation)RidgeR-NN · AT&Taccuracy: 90.86 -> 94.17Improved
add flexible Stiefel manifold SVM decision layer (ablation)RidgeR-NN · WAVEFORMaccuracy: 68.49 -> 84.26Improved
add flexible Stiefel manifold SVM decision layer (ablation)RidgeR-NN · UMISTaccuracy: 93.91 -> 97.39Improved
add adaptive sample weights to the SVM decision layer (ablation)RidgeR-SVM-NN · AT&Taccuracy: 94.17 -> 98.75Improved
add adaptive sample weights to the SVM decision layer (ablation)RidgeR-SVM-NN · WAVEFORMaccuracy: 84.26 -> 85.44Improved
add adaptive sample weights to the SVM decision layer (ablation)RidgeR-SVM-NN · UMISTaccuracy: 97.39 -> 97.39No clear effect
use non-gradient closed-form training instead of gradient descentDNN · UMIST and MNIST-Miniconvergence iterations: within 5 iterations vs more than 1000Improved
increase lambda above 2^0manifold neural network · AT&T and UMISTaccuracy and F1Hurt
vary lambda below 2^0manifold neural network · AT&T and UMISTaccuracy and F1No 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