Understanding human cognition through machine learning methods and its implications for explainable artificial intelligence

Speaker: Prof. Janet H. Hsiao
         Head and Associate Professor
         Department of Psychology
         University of Hong Kong

Title:  "Understanding human cognition through machine learning methods
         and its implications for explainable artificial intelligence"

Date:   Monday, 24 April 2023

Time:   4:00pm - 5:00pm

Venue:  Lecture Theater F (Leung Yat Sing Lecture Theater)
        near lift 25/26, HKUST

Abstract:

The cognitive revolution in the 1950s gave birth to the interdisciplinary
study of the mind later known as Cognitive Science, where computational
models of artificial intelligence (AI) provide essential scientific
methods for developing and testing theories of human cognition. In recent
years, rapid progress in AI has not only continuously provided powerful
methods to understand human cognition, but also raised important issues
about AI's comparability to human cognition. In this talk, I will
summarise some of my recent research in visual cognition to illustrate how
we can use machine learning methods to understand the representational and
computational capacities of the human mind, and how such understanding can
inform the development of explainable AI (XAI). I will first introduce a
machine learning based approach, Eye Movement analysis with Hidden Markov
Models (EMHMM), which provides quantitative measures of individual
differences in eye movements, taking both temporal and spatial dimensions
of eye movements into account. It has led to novel findings in cognitive
research not revealed by traditional analysis methods. By integrating it
with deep neural networks (DNN), we have developed DNN+HMM to account for
eye movement strategy learning in human visual cognition. I will then
introduce our recent studies comparing attention strategies in humans and
AI systems in object detection, and how these findings can inform XAI
design. These studies demonstrate the importance of interdisciplinary
approaches to the research on both human cognition and AI.


*****************
Biography:

Janet Hsiao is Head and Associate Professor in Department of Psychology, a
Principal Investigator of the State Key Laboratory of Brain and Cognitive
Sciences, and a Steering Committee member of Institute of Data Science at
University of Hong Kong. She received her Ph.D. in Informatics from
University of Edinburgh and was a postdoctoral researcher at University of
California San Diego. She is best known for her research on learning and
visual cognition, including face recognition, reading, and object
detection and identification. She adopts an interdisciplinary approach,
using methods and theories from artificial intelligence, experimental
psychology, psycholinguistics, and cognitive neuroscience to study the
human mind. In particular, she developed the Eye Movement analysis with
Hidden Markov Models (EMHMM) method to quantify individual differences in
eye movement pattern and consistency, which has revolutionized the use of
eye movement data to understand cognition. She has also combined EMHMM
with deep learning methods to understand the role of attention in human
learning and its implications for explainable Artificial Intelligence
research. She received the Best Language Modelling Paper Prize from the
Cognitive Science Society in 2006 and the Early Career Award from the RGC
Hong Kong in 2012. She is now a Fellow of the Cognitive Science Society
and serves on the Governing Board. She is also Editor-in-Chief of British
Journal of Psychology.