Susan A. Murphy
Bio:
Susan A. Murphy is a Professor at Harvard University where she leads the Statistical Reinforcement Learning Lab. Her research focuses on experimental designs and data analysis methods in sequential decision making with applications in health. She developed the "Sequential, Multiple Assignment Randomized Trial" for informing clinical decision making and the "Micro-Randomized Trial" for developing automated digital health interventions.
Abstract of her lecture to be held on 3 November 2022:
Personalization via Reinforcement Learning in Digital HealthIn health, a formidable challenge in designing the delivery of sequential treatments is to determine treatment policies: when and in which context it is best to deliver treatments. This is treatment personalization. Reinforcement Learning (RL) provides an attractive starting point for personalizing interventions. However, multiple challenges confront the use of RL in health, and in particular behavioral health care delivered via smart devices. This is a frontier area with primarily qualitative knowledge concerning the underlying system dynamics, delayed effects due to treatment burden, high noise and limited data. Further learning only on the current set of individuals wastes resources. The data should be useful for considering the causal effects of the decisions on a variety of auxiliary outcomes as well as to gather information to conduct studies to further optimize the treatment policy. A further challenge to RL is how to communicate uncertainty and confidence in the results. In this talk we discuss our ongoing efforts in striving to meet these challenges.
Marta Kwiatkowska
Bio:
Marta Kwiatkowska is a Professor in the Department of Computer Science and Fellow of Trinity College at the University of Oxford. Her research focuses on probabilistic modelling, automated verification and correct-by-construction synthesis techniques and tools, with applications to mobile robotics, distributed coordination, DNA computing and AI safety. She led the development of the PRISM probabilistic model checker (www.prismmodelchecker.org), which is widely used worldwide for research and teaching.
Abstract of her lecture to be held on 3 November 2022:
Probabilistic model checking for a safer world
Computing devices support us in almost all everyday tasks, from mobile phones and online banking to wearable medical devices and autonomous driving. Since embedded software at the heart of these devices must behave correctly in uncertain environments, probabilistic model checking techniques have been developed to guarantee their safety, reliability
and resource efficiency. Exemplified through the software tool PRISM
(prismmodelchecker.org), they have been successfully applied in a variety of domains, finding and fixing flaws in real-world systems. Using illustrative examples, this lecture will give an overview of the role that probabilistic modelling, verification and synthesis can play
in a variety of applications, including robotics, security, medical devices and DNA computing. Finally, challenges in verification of systems that include machine learning components will be outlined, and how these may be overcome.
