Abstract
Gaze Trajectories (GTs) are important symptoms of certain health conditions, namely psychiatric and neurological diseases. In this talk, we aim at showing new mathematical and Machine Learning (ML) methods to classify GTs and identify signatures of health conditions.
While many studies simplify eye-movements by using top-down processing tasks, such as looking left or right upon certain cues, in this project we aim instead at maximizing the information GTs can reveal. This is done, we hypothesize, when the eyes can move freely in exploring an image. In this context, we assume that GTs have the ultimate goal of maximizing information gathered in a given period, or, in other words, that individuals are engaged in foraging for visual information. Here we talk why this hypothesis is relevant to the field of eye-tracking and provide a solution to this problem.
We will also cover some AI methods to analyse gaze trajectories and their limitations in doing so. Namely, we focus on GANs with a double purpose. If they were capable of faithfully replicating GTs, not only could we augment our data or create faithful synthetic data with less ethical constrains, but we could also use the discriminator part of the GAN to classify gaze trajectories. We apply some of these results to create a classification of GTs typical of autistic children.
Finally, we cover some current and future research activities at OsloMet.
About the speaker
Pedro Lencastre is PhD candidate at OsloMet in the field of Intelligent Health, with background in Physics and Mathematics. He conducts research in the field of eye-tracking, namely on mathematical and AI methods of time-series classifications.