-------------

Abstract topic: Collective Online Learning of Gaussian Processes in Massive Multi-Agent Systems

Distributed machine learning (ML) is a modern computation paradigm that divides its workload into independent tasks that can be simultaneously achieved by multiple machines (i.e., agents) for better scalability. However, a typical distributed system is usually implemented with a central server that collects data statistics from multiple independent machines operating on different subsets of data to build a global analytic model. This centralized communication architecture however exposes a single choke point for operational failure and places severe bottlenecks on the server's communication and computational capacities as it has to process a growing volume of communication from a crowd of learning agents. In this talk, I will introduce a novel collective learning framework for massively distributed systems that allow each agent to build its local predictive model, which can be exchanged and combined efficiently with others via peer-to-peer communication to converge on a global model of higher quality. I will also present some empirical results that consistently demonstrate the efficiency of the proposed framework on several datasets.

-------------

Abstract topic: Collective Online Learning of Gaussian Processes in Massive Multi-Agent Systems

Due to the recent development of novel technologies such as the Internet of Things (IoT), autonomous machines (e.g., driverless cars, or UAVs), and crowdsourcing, human-aware AI systems (i.e., systems 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. In many cases, the required objectives of these collectives can be achieved via making a sequence of decisions under uncertainty. However, standard solution techniques for these types of problems fail to efficiently tackle these objectives, as they typically ignore the budget constraints caused by the human factor.  For example, algorithms that require a sequence of iterative optimization steps (e.g., gradient descent) assume that the participating agent (or human) will follow the steps of the optimization process, described by the algorithm. However, humans might get disengaged or unmotivated, and thus, might leave the process, or intentionally not follow the instructions given by the algorithm. Another example comes from the potential miscommunication between humans and agents, typically caused by inefficient interaction interfaces. Such new constraints introduce new challenges in solving sequential decision-making problems of human-aware AI systems. Whereas some solutions have been proposed to tackle specific sub-problems with restricted settings, no effort has been made to date to investigate the problems of user motivation from a more fundamental and generic perspective, which is essential to fully understand these problems, and thus, to provide more efficient solutions to tackle them. In this talk, I will discuss some research solutions within each above-mentioned topic. In particular, I will mainly focus on tackling budgeted human factors in sequential decision-making problems, a research area I have been working on in the last few years.

Sincerely, 

#TalkSeries2019