01 March 2021 - Arrasy Rahman, Jorge Gaete, Jonathan Feldstein

 

Speaker: Jonathan Feldstein

 

Title: Lifted Reasoning meets Weighted Model Integration

 

Abstract:

Exact inference in probabilistic graphical models is particularly challenging in the presence of relational and other deterministic constraints. For discrete domains, weighted model counting has emerged as an effective and general approach in a variety of formalisms, exploiting symmetry properties over indistinguishable groups of objects in a relational setting, known as lifted reasoning has pushed the envelope further, by avoiding the need to perform inference on the exponential ground theory.

 

Given the limitation to discrete domains, the formulation of weighted model integration was proposed as an extension to weighted model counting for mixed discrete-continuous domains over both symbolic and numeric weight functions. While that formulation has enjoyed considerable attention in recent years, there is very little understanding on whether the task can be solved at a lifted level, that is, whether we can reason with relational models by avoiding grounding. In this paper, we consider this question. We show how to generalize algorithmic ideas known in the circuit compilation for function-free lifted inference to functions with a continuous range.

 

Bio:

Jonathan is a second year PhD student, supervised by Vaishak Belle and James Cheney. He is interested in neuro-symbolic AI and lifted reasoning with possible applications in privacy and security.

 

 

Speaker: Arrasy Rahman

 

Title: Towards Open Ad Hoc Teamwork Using Graph-based Policy Learning

 

Abstract:

Ad hoc teamwork is the challenging problem of designing an autonomous agent that can adapt quickly to collaborate with teammates without prior coordination mechanisms, including joint training. Prior work in this area has focused on closed teams in which the number of agents is fixed. In this work, we consider open teams by allowing agents with different fixed policies to enter and leave the environment without prior notification. Our solution builds on graph neural networks to learn agent and joint-action value models under varying team compositions. We contribute a novel action-value computation that integrates the agent and joint-action value model to produce action-value estimates. We empirically demonstrate that our approach successfully models the effects other agents have on the learner, which leads to policies that robustly adapt to dynamic team compositions and significantly outperform several alternative methods.

 

Bio:

Arrasy is a third year PhD student supervised by Stefano Albrecht. He is interested in relational reinforcement learning with Graph Neural Networks (GNNs) and its applications to multiagent systems.

 

 

Speaker:  Jorge Gaete

 

Title: Clustering approaches to understand multimorbidity patterns

 

Abstract:

Multimorbidity, the co-occurrence of more than one illness, is an increasing condition worldwide as the population ages. This condition is associated with a decrease in patient’s quality of life, negative medical outcomes, and impacts on the utilization of medical resources. Assessing patient’s multimorbidities is particularly challenging as medical protocols for treatment and diagnosis are usually designed under the assumption of a single morbidity per patient. In this talk we present our explorations of clustering techniques to understand patterns of multimorbidities, examine classic approaches such as k-means as well as network approaches, and discuss some of our current findings and challenges for future work.

 

Bio:

Jorge is a second year PhD student supervised by Jacques Fleuriot. He is interested in explainable AI for the healthcare domain, especially in the intersection between machine learning and symbolic AI.