Post

Convergence in Off-Policy Settings

A Brief Convergence Analysis of Existing Algorithms in the Literature

⚠️ Under construction!

Off-policy reinforcement learning (RL) presents a challenging yet essential setting for the practical application of RL algorithms in real-world scenarios. In this post, we provide a concise analysis of existing algorithms in both off-policy and offline settings.

Post

Optimization to Reinforcement Learning

An Optimization Perspective on Reinforcement Learning (RL) for Algorithm Design

⚠️ Under review!

Drawing inspiration from the correspondence between optimization and Markov Decision Processes (MDPs), we explore extensions of existing optimization algorithms to RL, employing a mathematically rigorous approach. In particular, we investigate Quasi-Newton methods and propose a novel RL algorithm that demonstrates convergence in both model-based and model-free settings, achieving memory and computational complexity.