BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Scaling Laws and Spectra of Shallow Neural Networks in the Feature
 -Learning Regime
DTSTART;VALUE=DATE-TIME:20260416T210000Z
DTEND;VALUE=DATE-TIME:20260416T213000Z
DTSTAMP;VALUE=DATE-TIME:20260413T221135Z
UID:indico-contribution-358@fisindico.uniandes.edu.co
DESCRIPTION:Speakers: Bruno Loureiro (ENS Paris)\nNeural scaling laws unde
 rlie many of the recent advances in deep learning\, yet their theoretical 
 understanding remains largely confined to linear models. In this work\, we
  present a systematic analysis of scaling laws for quadratic and diagonal 
 neural networks in the feature learning regime. Leveraging connections wit
 h matrix compressed sensing and LASSO\, we derive a detailed phase diagram
  for the scaling exponents of the excess risk as a function of sample comp
 lexity and weight decay. This analysis uncovers crossovers between distinc
 t scaling regimes and plateau behaviours\, mirroring phenomena widely repo
 rted in the empirical neural scaling literature. Furthermore\, we establis
 h a precise link between these regimes and the spectral properties of the 
 trained network weights\, which we characterize in detail. Consequently\, 
 we provide a theoretical validation of recent empirical observations conne
 cting the emergence of power-law tails in the weight spectrum with network
  generalization performance\, yielding an interpretation from first princi
 ples.\n\nhttps://arxiv.org/abs/2509.24882\n\nhttps://fisindico.uniandes.ed
 u.co/event/23/contributions/358/
LOCATION:Universidad de los Andes ML-513
URL:https://fisindico.uniandes.edu.co/event/23/contributions/358/
END:VEVENT
END:VCALENDAR