A Continuous Information Gain Measure to Find the Most Discriminatory Problems for AI Benchmarking

TL;DR

This paper presents an information-theoretic approach to select a minimal set of problems that maximizes discriminatory power among AI algorithms, improving benchmarking efficiency in game environments.

Contribution

It introduces a continuous information gain measure for problem selection, enabling efficient identification of the most informative problems for AI benchmarking.

Findings

A small subset of games retains most discriminatory information.

The method reduces computational costs in AI benchmarking.

It reveals which games best differentiate between different AI agents.

Abstract

This paper introduces an information-theoretic method for selecting a subset of problems which gives the most information about a group of problem-solving algorithms. This method was tested on the games in the General Video Game AI (GVGAI) framework, allowing us to identify a smaller set of games that still gives a large amount of information about the abilities of different game-playing agents. This approach can be used to make agent testing more efficient. We can achieve almost as good discriminatory accuracy when testing on only a handful of games as when testing on more than a hundred games, something which is often computationally infeasible. Furthermore, this method can be extended to study the dimensions of the effective variance in game design between these games, allowing us to identify which games differentiate between agents in the most complementary ways.