The AAAI Conference on Artificial Intelligence will be taking place on February 7-12, 2020 at the Midtown Manhattan Hilton. Researchers in AI from around the world will gather to discuss recent breakthroughs and define topics of interest for the coming year.

SigOpt is excited to participate in the AI job fair at AAAI taking place on February 11, 2020 from 12:30 – 3:00 PM. Our Research Engineers, Michael McCourt and Gustavo, will be attending the conference. If you are interested to learn more about SigOpt, San Francisco, Silicon Valley or startup life, stop by the career fair booth and chat with our team.

Accelerating Psychometric Screening Tests with Prior Information

Among the researchers at AAAI is SigOpt research engineer Gustavo Malkomes, who will be presenting his latest research at the Health Intelligence workshop. The article is a continuation of joint research between Gustavo at Dr. Dennis Barbour’s research team at the Washington University of St. Louis; earlier results were presented at the NeurIPS workshop on ML for Health.

Abstract

Classical methods for psychometric function estimation either require excessive measurements or produce only a low-resolution approximation of the target psychometric function. In this paper, we propose two solutions for rapid high-resolution approximation of the psychometric function of a patient given her or his prior exam.

First, we develop a rapid screening algorithm for a change in the psychometric function estimation of a patient. We use Bayesian active model selection to perform an automated pure-tone audiometry test with the goal of quickly finding if the current estimation will be different from the previous one. Second, we propose an algorithm for learning the patient’s audible threshold. We use the previous test as a prior belief to reduce the number of measurements necessary for estimation.

We validate our methods using audiometric data from the National Institute for Occupational Safety and Health (NIOSH). Initial results indicate that with a few tones we can i) detect if the patient’s audiometric function has changed between the two test sessions with high confidence, and ii) learn high-resolution approximations of the target psychometric function.