Machine Learning with Noisy Labels

Chapter 5: Benchmarks, methods, results, and code

Abstract

In this chapter we present the main noisy-label learning benchmarks and state-of-the-art methods being studied in the field. These benchmarks form a subset of the challenges presented in Chapter 2, and they are used to compare the performance of methods with the goal of understanding which techniques are more effective to solve particular noisy-label learning problems. We also show the top-performing methods per benchmark, where we aim to list all techniques (from Chapter 4) that each method explores. We conclude the chapter by summarizing the most explored techniques by the top-performing methods.

Keywords

Noisy-label learning techniques; Noisy-label learning benchmark; Noisy-label learning ...