Decision Awareness in Reinforcement Learning - End-to-End Optimization Approaches
GERAD Research Center via YouTube
Overview
Syllabus
Intro
Definition
End-to-End Principle
What is neural network?
Automatic Differentiation
Computational Graph
Reverse and Forward Mode
Control-Oriented Model-Based Reinforcement Learning with Implicit Differentiation
Optimal Model Design Problem (OMD)
Smooth Bellman Optimality Equations
Connection between OMD and Rust (1988)
Bilevel Optimization (Bard 1998)
Implicit and Iterative Differentiation
Benefits under Model Misspecification
Function Approximation and Distractor States
Performance under Model Misspecification
Continuous-Time Meta-Learning with Forward Mode Dif- ferentiation.
Gradient Flow-based Meta-Learning
Time irreversibility
Memory-efficient meta-gradients
Consequence
Empirical Efficiency of COML
Nonlinear Trajectory Optimization
Extragradient Method
Trajectory Optimization with Learned Model
Taught by
GERAD Research Center