Skip to main navigation Skip to search Skip to main content

Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training

  • Max Milkert
  • , David Hyde
  • , Forrest Laine
  • Vanderbilt University

Research output: NLRPoster

Abstract

Deep ReLU networks have exponential expressive potential but behave identically to their shallow counterparts under random initialization. This paper introduces a novel training paradigm that constrains weights into mathematically optimized patterns. This enables the resulting network to make exponential use of its depth.
Original languageAmerican English
PublisherNational Laboratory of the Rockies (NLR)
Number of pages1
DOIs
StatePublished - 2025

Publication series

NamePresented at the International Conference on Machine Learning (ICML), 13-19 July 2025, Vancouver, Canada

NLR Publication Number

  • NLR/PO-2C00-95812

Keywords

  • activation regions
  • linear regions
  • network initialization
  • ReLU activation

Fingerprint

Dive into the research topics of 'Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training'. Together they form a unique fingerprint.

Cite this