Thursday 10:45–11:30 in Track 1, Track 2, Track 3

Towards Interpretable Accountable Models

Katharine Jarmul

Audience level:
Intermediate

Description

In a world where machine learning can affect human lives in unprecedented ways, how can we create interpretable and accountable models? How do we then share the reasonings with our colleagues? This talk will review why interpretability is key in public facing models as well as tools and methods for creating accurate and explainable models.

Abstract

Machine learning might be used to determine if you get a job interview, a bank loan or an acceptance letter at a university. In a world where machine learning can affect human lives in unprecedented ways, how can we create interpretable and accountable models? Once we can interpret the model used, how do we then share the reasoning with our colleagues?

This talk will review why interpretability is key in public facing models as well as tools and methods for creating accurate and explainable models. We will review the newest publications on increasing accountability in model training and see techniques for building understandable models from more difficult architectures (such as deep learning).

Subscribe to Receive PyData Updates

Tickets

Get Now