Publications

Sorted by DateClassified by Publication TypeClassified by Research Category

Loss Bounds for Approximate Influence-Based Abstraction

Elena Congeduti, Alexander Mey, and Frans A. Oliehoek. Loss Bounds for Approximate Influence-Based Abstraction. In Proceedings of the Twentieth International Conference on Autonomous Agents and Multiagent Systems (AAMAS), pp. 377–385, May 2021.

Download

pdf [1.5MB]  

Abstract

Sequential decision making techniques hold great promise to improve the performance of many real-world systems, but computational complexity hampers their principled application. Influence-based abstraction aims to gain leverage by modeling local subproblems together with the ‘influence’ that the rest of the system exerts on them. While computing exact representations of such influence might be intractable, learning approximate representations offers a promising approach to enable scalable solutions. This paper investigates the performance of such approaches from a theoretical perspective. The primary contribution is the derivation of sufficient conditions on approximate influence representations that can guarantee solutions with small value loss. In particular we show that neural networks trained with cross entropy are well suited to learn approximate influence representations. Moreover, we provide a sample based formulation of the bounds, which reduces the gap to applications. Finally, driven by our theoretical insights, we propose approximation error estimators, which empirically reveal to correlate well with the value loss.

BibTeX Entry

@inproceedings{Congeduti21AAMAS,
    author= {Elena Congeduti and
            Alexander Mey and
            Frans A. Oliehoek},
    title =     {Loss Bounds for Approximate Influence-Based Abstraction},
    booktitle = AAMAS21,
    year =      2021,
    month =     may,
    pages =     {377--385},
    keywords =   {refereed},
    abstract = {
Sequential decision making techniques hold great promise to improve the performance of many real-world systems, but computational complexity hampers their principled application. Influence-based abstraction aims to gain leverage by modeling local subproblems together with the ‘influence’ that the rest of the system exerts on them. While computing exact representations of such influence might be intractable, learning approximate representations offers a promising approach to enable scalable solutions. This paper investigates the performance of such approaches from a theoretical perspective. The primary contribution is the derivation of sufficient conditions on approximate influence representations that can guarantee solutions with small value loss. In particular we show that neural networks trained with cross entropy are well suited to learn approximate influence representations. Moreover, we provide a sample based formulation of the bounds, which reduces the gap to applications. Finally, driven by our theoretical insights, we propose approximation error estimators, which empirically reveal to correlate well with the value loss.
    }
} 

Generated by bib2html.pl (written by Patrick Riley) on Wed Jun 23, 2021 12:23:25 UTC