Skip to main navigation Skip to search Skip to main content

SIAM Op26 Minisymposium: Implicit Bias in Neural Network Optimization

Activity: Academic conferences and events (excluding conference publications) › Major conference organising role

Description

Implicit bias in neural-network optimization is central to understanding generalization, robustness, and training dynamics. Although all optimizers minimize empirical risk, different algorithms favor different kinds of solutions. This minisymposium surveys recent advances on optimizer bias: how it emerges in large-scale nonconvex systems, how to characterize it spectrally and geometrically, and what it implies for theory, generalization, and practice. Talks will draw on optimization, random matrix theory, statistics, and high-dimensional geometry, with applications to overparameterized models, deep networks, and structured data. Four sessions will bring together a research community at the optimization–ML interface, balancing foundations, algorithmic insights and applications to spark discussion and identification of high value open questions.
Period3 May 2026
Event typeWorkshop
LocationEdinburgh, UK United KingdomShow on map
Degree of RecognitionInternational