A Markovian Model for Learning-to-Optimize

Name: A Markovian Model for Learning-to-Optimize
Uploaded: 2026-08-20T13:33:11.354363+00:00
Duration: 737 s
Description: The episode discusses 'A Markovian Model for Learning-to-Optimize,' which provides a rigorous probabilistic framework for training optimization algorithms. Hosts explain how the paper uses Markov chains and PAC-Bayesian bounds to provide high-probability guarantees on convergence time and rate, moving learned optimizers from black boxes to provable tools.

The episode discusses 'A Markovian Model for Learning-to-Optimize,' which provides a rigorous probabilistic framework for training optimization algorithms. Hosts explain how the paper uses Markov chains and PAC-Bayesian bounds to provide high-probability guarantees on convergence time and rate, moving learned optimizers from black boxes to provable tools.

Open the full episode