Re-evaluating Evaluation

TL;DR

The paper introduces Nash averaging as a new evaluation method for machine learning that adapts to redundancies, promoting inclusive and unbiased assessment across agents and tasks.

Contribution

It proposes Nash averaging, a novel evaluation framework based on game theory, to address biases caused by task and agent redundancies in machine learning assessments.

Findings

Nash averaging automatically adjusts for redundancies in evaluation data.

It promotes inclusive evaluation by mitigating biases from easy tasks or weak agents.

The approach is grounded in algebraic analysis of agent-vs-agent and agent-vs-task scenarios.

Abstract

Progress in machine learning is measured by careful evaluation on problems of outstanding common interest. However, the proliferation of benchmark suites and environments, adversarial attacks, and other complications has diluted the basic evaluation model by overwhelming researchers with choices. Deliberate or accidental cherry picking is increasingly likely, and designing well-balanced evaluation suites requires increasing effort. In this paper we take a step back and propose Nash averaging. The approach builds on a detailed analysis of the algebraic structure of evaluation in two basic scenarios: agent-vs-agent and agent-vs-task. The key strength of Nash averaging is that it automatically adapts to redundancies in evaluation data, so that results are not biased by the incorporation of easy tasks or weak agents. Nash averaging thus encourages maximally inclusive evaluation -- since there is no harm (computational cost aside) from including all available tasks and agents.