A Survey of Learning Criteria Going Beyond the Usual Risk

TL;DR

This paper surveys alternative learning criteria beyond traditional average loss, emphasizing the importance of desirable loss distributions and providing historical context and new perspectives on machine learning evaluation.

Contribution

It introduces a broad range of non-traditional criteria for designing and evaluating algorithms, challenging the conventional focus on expected loss.

Findings

Highlights limitations of average loss optimization

Proposes viewing learning through the lens of loss distribution quality

Provides historical context and new perspectives on evaluation criteria

Abstract

Virtually all machine learning tasks are characterized using some form of loss function, and "good performance" is typically stated in terms of a sufficiently small average loss, taken over the random draw of test data. While optimizing for performance on average is intuitive, convenient to analyze in theory, and easy to implement in practice, such a choice brings about trade-offs. In this work, we survey and introduce a wide variety of non-traditional criteria used to design and evaluate machine learning algorithms, place the classical paradigm within the proper historical context, and propose a view of learning problems which emphasizes the question of "what makes for a desirable loss distribution?" in place of tacit use of the expected loss.