Breaking Down the Inductive Proofs Behind Faster Value Iteration in RL

Written by anchoring | Published 2025/01/15
Tech Story Tags: reinforcement-learning | dynamic-programming | nesterov-acceleration | machine-learning-optimization | value-iteration | value-iteration-convergence | bellman-error | bellman-convergence-analysis

TLDRExplore how Anchored Value Iteration (Anc-VI) accelerates convergence in dynamic programming and reinforcement learningvia the TL;DR App

Authors:

(1) Jongmin Lee, Department of Mathematical Science, Seoul National University;

(2) Ernest K. Ryu, Department of Mathematical Science, Seoul National University and Interdisciplinary Program in Artificial Intelligence, Seoul National University.

Abstract and 1 Introduction

1.1 Notations and preliminaries

1.2 Prior works

2 Anchored Value Iteration

2.1 Accelerated rate for Bellman consistency operator

2.2 Accelerated rate for Bellman optimality opera

3 Convergence when y=1

4 Complexity lower bound

5 Approximate Anchored Value Iteration

6 Gauss–Seidel Anchored Value Iteration

7 Conclusion, Acknowledgments and Disclosure of Funding and References

A Preliminaries

B Omitted proofs in Section 2

C Omitted proofs in Section 3

D Omitted proofs in Section 4

E Omitted proofs in Section 5

F Omitted proofs in Section 6

G Broader Impacts

H Limitations

B Omitted proofs in Section 2

First, we prove the following lemma by induction.

Now, we present our key lemmas for the first rate of Theorem 2.

and let U¯ be the entire right hand side of inequality. Then, we have

By induction,

and let U¯ be the entire right hand side of inequality. Then, we have

Now, we prove the first rate of Theorem 2.

where the second inequality is from the condition.

By induction,

and let U¯ be the entire right hand side of inequality. Then, we have

Now, we prove the second rates of Theorem 2.

This paper is available on arxiv under CC BY 4.0 DEED license.


Written by anchoring | Anchoring provides a steady start, grounding decisions and perspectives in clarity and confidence.
Published by HackerNoon on 2025/01/15