Interpretable Machine Learning 论文

摘要

Interpretable machine learningThis POSTnote gives an overview of machine learning (ML) and its role in decision-making.It examines the challenges of understanding how a complex ML system has reached its output, and some of the technical approaches to making ML easier to interpret.It gives a brief overview of some of the proposed tools for making ML systems more accountable, such as algorithm audit and impact assessments. Overview◼ Machine learning (ML) is being used to support decision-making in applications such as recruitment and medical diagnoses.◼ Concerns have been raised about some complex types of ML, where it is difficult to understand how a decision has been made.◼ A further risk is the potential for ML systems to introduce or perpetuate biases.◼ Approaches to improving the interpretability of ML include designing systems using simpler methods and using tools to gain an insight into how complex systems function.◼ Interpretable ML can improve user trust and ML performance, however there are challenges such as commercial sensitivity.◼ Proposed ways to improve ML accountability include auditing and impact assessments.