Eric Nalisnick - On the Calibration of Learning-to-Defer Systems
|Starts:||14:00 9 Feb 2022|
|Ends:||15:00 9 Feb 2022|
|What is it:||Seminar|
|Organiser:||Department of Mathematics|
|Who is it for:||University staff, External researchers, Current University students|
Eric Nalisnick, Assistant Professor at the Amsterdam Machine Learning Lab, Informatics Institute at the University of Amsterdam is our speaker for the Statistics seminar series.
Title: On the Calibration of Learning-to-Defer Systems
Abstract: Learning to defer (L2D) systems offer a promising solution to the problem of AI safety. For a given input, the system can defer the decision to a human if the human is more likely than the model to take the correct action. We study the calibration of L2D systems, investigating if the probabilities they output are sound. We find that Mozannar & Sontag’s (2020) multiclass framework is not calibrated with respect to expert correctness. Moreover, it is not even guaranteed to produce valid probabilities due to its parameterization being degenerate for this purpose. We propose an L2D system based on one-vs-all classifiers that is able to produce calibrated probabilities of expert correctness. Furthermore, our loss function is also a consistent surrogate for multiclass classification, like Mozannar & Sontag’s (2020). In addition to being calibrated, our model exhibits accuracy that is comparable to Mozannar & Sontag’s (2020) model (and often better) in tasks ranging from hate speech detection to galaxy classification to diagnosis of skin lesions.
Organisation: University of Amsterdam
Travel and Contact Information
Zoom link: https://zoom.us/j/92947173491