This document summarizes a presentation on offline reinforcement learning. It discusses how offline RL can learn from fixed datasets without further interaction with the environment, which allows for fully off-policy learning. However, offline RL faces challenges from distribution shift between the behavior policy that generated the data and the learned target policy. The document reviews several offline policy evaluation, policy gradient, and deep deterministic policy gradient methods, and also discusses using uncertainty and constraints to address distribution shift in offline deep reinforcement learning.
This document proposes a method called Fast DiffusionMBIR to solve 3D inverse problems using pre-trained 2D diffusion models. It augments the 2D diffusion prior with a model-based total variation prior to encourage consistency across image slices. The method performs denoising across image slices in parallel using a 2D diffusion score function, and then jointly optimizes data consistency and the total variation prior between slices. It shares primal and dual variables between iterations for faster convergence. Results on sparse-view CT reconstruction show coherent volumetric results across all slices.
This document proposes a method called Fast DiffusionMBIR to solve 3D inverse problems using pre-trained 2D diffusion models. It augments the 2D diffusion prior with a model-based total variation prior to encourage consistency across image slices. The method performs denoising across image slices in parallel using a 2D diffusion score function, and then jointly optimizes data consistency and the total variation prior between slices. It shares primal and dual variables between iterations for faster convergence. Results on sparse-view CT reconstruction show coherent volumetric results across all slices.