convergence properties
Convergence properties characterize how a learning algorithm approaches a solution over iterations, including aspects like speed and stability, which are essential for ensuring effective training of models.
- Accelerating Model-Free Optimization via Averaging of Cost Samples
- Mind the Gap: Removing the Discretization Gap in Differentiable Logic Gate Networks
- New Perspectives on the Polyak Stepsize: Surrogate Functions and Negative Results
- Nonlinearly Preconditioned Gradient Methods: Momentum and Stochastic Analysis
- Preference Learning with Response Time: Robust Losses and Guarantees
- Revisiting 1-peer exponential graph for enhancing decentralized learning efficiency
- Semi-infinite Nonconvex Constrained Min-Max Optimization
- Sharp Gaussian approximations for Decentralized Federated Learning
- U-REPA: Aligning Diffusion U-Nets to ViTs