Modelling and influencing behaviour using recurrent neural networks and reinforcement-learning.
Date: 27/10/21 13.00-13.50 AEDT
Speaker: Dr Amir Dezfouli https://adezfouli.github.io/
Title: Modelling and influencing behaviour using recurrent neural networks and reinforcement-learning.
Bio: I earned my BSc and MSc in software engineering and artificial intelligence from the University of Tehran. I then moved to Australia and earned my PhD from the University of Sydney, in which I studied computational mechanisms of hierarchical decision-making. I then spent two years at the University of New South Wales (UNSW) working on deep neural network models of behavioural and neural data. I joined the machine learning research group in CSIRO, Sydney in 2018, and since then my research has been centered around machine learning and studying artificial and biological decision-making systems.
Abstract: It has recently been suggested that recurrent neural networks provide a flexible way for modelling human behaviour. In this talk, I will present different recurrent neural network architectures for modelling behavioural data and compare their abilities with traditional computational models. I’ll further present a reinforcement-learning framework in which recurrent neural networks can be used to influence human choices and explore its ability in simple scenarios to complex social decision-making tasks.