Program for stream Optimization in neural architectures: convergence and solution characterization

WC-03: Optimization in neural architectures I
Stream: Optimization in neural architectures: convergence and solution characterization
Room: M:J
Chair(s): Manish Krishan Lal, Maria-Luiza Vladarean

• Optimization dynamics of equivariant neural networks
  Axel Flinth
• Understanding neural architectures via projection onto sets of generalized bilinear constraints
  Manish Krishan Lal

WD-03: Optimization in neural architectures II
Stream: Optimization in neural architectures: convergence and solution characterization
Room: M:J
Chair(s): Maria-Luiza Vladarean, Manish Krishan Lal

• Vanishing Gradients in Reinforcement Finetuning of Language Models
  Noam Razin
• A phase transition between positional and semantic learning in a solvable model of dot-product attention
  Hugo Cui, Freya Behrens, Florent Krzakala, Lenka Zdeborová
• On the spectral bias of two-layer linear networks
  Aditya Varre