Machine Learning with Noisy Labels

Chapter 2: Noisy-label problems and datasets

Tasks, taxonomy, datasets, and evaluation

Abstract

This chapter starts with a definition of intrinsic label noise, followed by explanation of the main types of label noise, namely: symmetric, asymmetric, and instance-dependent noise. We also provide definitions of closed-set and open-set label noise problems. We then provide a detailed list of current datasets and benchmarks available in the field to assess label noise algorithms and models. The chapter is concluded with a presentation of the main evaluation methods used to assess the performance of approaches designed to address label noise problems.

Keywords

Intrinsic label noise; Symmetric label noise; Asymmetric label noise; Instance-dependent label ...