Optimal Gradient-Based Algorithms for Non-Concave Bandit Optimization

Optimal Gradient-Based Algorithms for Non-Concave Bandit Optimization

Explore a 31-minute lecture by Qi Lei from Princeton University on optimal gradient-based algorithms for non-concave bandit optimization. Delve into advanced topics including the stochastic bandit eigenvector problem, noisy power methods, and stochastic low-rank linear rewards. Examine information theoretical understanding, regret comparisons for quadratic rewards, and higher-order polynomial bandit problems. Learn about optimal regret for different types of reward functions and potential extensions to reinforcement learning in simulator settings. Gain insights into cutting-edge research in sampling algorithms and geometries on probability distributions presented at the Simons Institute.