AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

On the Stability and Scalability of Node Perturbation Learning

Naoki Hiratani, Yash Mehta, Timothy P Lillicrap, Peter E. Latham

NeurIPS 2022 · Optimization › Implicit bias and generalization · approach: Neural tangent kernel analysis

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
increase number of output units Ly in node perturbation (NP) versus SGDNP, linear regression · linear regressiontraining time to target errorHurt
increase hidden layer size Lh for NP in one-hidden-layer linear networksNP · one-hidden-layer linear networkminimum training timeNo clear effect
increase output layer size Ly for NP in one-hidden-layer linear networkNP · one-hidden-layer linear networkminimum training timeHurt
raise learning rate above critical value eta*NP with teacher noise · one-hidden-layer linear networkiterations until error starts to increase: eta^-1.5 -> eta^-3 scalingHurt
add weight normalization (constant incoming weight norm per neuron) (ablation)NP · one-hidden-layer linear network with teacher noiseconvergence stabilityMixed
increase hidden layer width for NP vs SGD weight update alignmentNP · SARCOScosine similarity with SGD in hidden layer: falls off as 1/sqrt(Lh)Hurt
increase learning rate for NPNP · SARCOStest error stabilityHurt
use SGD instead of NP regarding late instabilitySGD · SARCOSlate-training instabilityImproved

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