Early-Signal Detector

method · first seen 8 Oct 2026

gradient flow

The continuous-time limit of gradient descent used to analyze how neural network training dynamics evolve.

Mentions per week

0119 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

stochastic gradient descentmultilayer perceptrontwo-layer ReLU network teacher-student modelsymmetry breaking in neural network trainingregularizationJKO scheme