June 6, 2024 09:19
Talk by Prof. Richard Zemel (Columbia University, USA) </br>on May 28, 2024 thumbnails


Date and Time:
May 28, 2024: 10:30 am – 11:30 am (JST)
Venue: Online and Open Space at the RIKEN AIP Nihonbashi office

TITLE: Quantile Risk Control: A Framework for Responsible Deployment of AI Models

SPEAKER: Prof. Richard Zemel (Columbia University, USA)

Explicit statistical guarantees on model performance are an important ingredient in responsible machine learning. I will describe a simple and flexible framework that allows us to construct rigorous bounds on the loss of deployed prediction models. We have developed the framework with the aim of applying it to a number of areas. One area is algorithmic fairness, where it can control the extent to which different members of a population experience unequal effects of algorithmic decisions. A second application area is pretrained language models, where the framework helps determine how best to prompt a model to perform a given task. I will also describe extensions of the framework, to accommodate the possibility of distribution shifts in deployment. Experiments on applications such as chatbots, medical question summarization, and code generation highlight how such a framework can reduce the risk of the worst outcomes.

RIKEN AIP Distinguished Lecture Lecture Series URL: