Optimisation Problems in Human-Agent Learning - Long Tran-Thanh

Speaker: Dr. Long Tran-Thanh Associate Professor, University of Warwick Date: 24th June 2021 Title: Optimisation Problems in Human-Agent Learning Abstract: Due to the recent development of novel technologies such as Internet of Things (IoT), autonomous machines (e.g., driverless cars, or UAVs), and crowdsourcing systems, human-agent collectives, in which humans and machines (or agents) have to perform some degree of cooperation with each other in order to achieve certain goals, are becoming a reality. These collectives oftentimes need to perform a number of machine learning tasks, which can be formulated as online optimisation problems. However, standard optimisation solution techniques fail to efficiently tackle these objectives, as they typically ignore the optimisation constraints caused by the human factor. For example, algorithms that require a sequence of iterative optimisation steps (e.g., gradient descent) assume that the participating agent (or human) will follow the steps of the optimisation process, described by the algorithm. However, humans might get disengaged or unmotivated, and thus, might leave the process, or intentionally do not follow the instructions given by the algorithm (e.g., due to some selfish/strategic reasons). Such new constraints introduce new challenges in solving optimisation problems of human-agent collectives. In this talk, I will discuss some approaches for a number of optimisation problems with human factors, including: (i) incentive engineering (i.e., how to motivate human users to participate in the process); (ii) online optimisation/learning with strategic agents (e.g., when humans/agents intentionally do not follow our instructions); and (iii) secure human-agent learning. Finally, I will also discuss how to apply these findings in AI for Social Good.