Long-Term Real-World Personalization for Autonomous Systems

Abstract:

As autonomous systems are increasingly integrated into daily life, a key question is how to ensure personalization to individual preferences and unique environments of diverse users. A central challenge is that over the long term, each user's preferences and environments can change over time, and it is impossible to predict these changes at design time, thus rendering pre-trained general-purpose systems impractical for personalization. In this talk, I postulate that because humans are the ones best aware of their dynamic needs, they are perfectly suited to communicate these aspects to the system. To this end, I will discuss the main challenges addressed by my prior research and future directions towards long-term personalization through human-interactive continual robot learning. I will discuss my insights that effective learning from non-expert human input requires robot-centered aspects, such as scalability, robustness, and computational efficiency, as well as human-centred aspects, such as interpretability, fairness, and trustworthiness. 

Biography:

Dr. Ali Ayub is an Assistant Professor at Concordia Institute for Information Systems Engineering (CIISE), Concordia University where he directs the PaInt (Personal and Interactive Autonomous Systems) lab. Before joining Concordia, he was a postdoctoral fellow at the University of Waterloo with Kerstin Dautenhahn and Chrystopher Nehaniv. He received his MS and PhD in Electrical Engineering from Penn State in 2017 and 2021, respectively, with Alan Wagner. His research interests include personalization, interactive machine learning, lifelong learning, and assistive robotics. He aims to develop autonomous systems that can continually adapt to individual user preferences and unique environments to provide personalized assistance over the long term.