Warm starting adaptive interventions with side-information from confounded logs
When
This talk focuses on adaptive interventions, where we have access to logged data on the effects of potential interventions, and our objective is to use this data to warm-start future interventions. In our setting, the logged data has been collected using a policy that had access to rich contextual information, which has however only been partially recorded (aka partially confounded logs). Such problems appear in a variety of settings including health interventions and online advertising/recommendations. We present UCB-SI, a variant of the classic UCB algorithm for navigating explore-exploit trade-offs. UCB-SI uses upper and lower bounds on the effects of potential interventions (derived from the partially confounded logs) to adaptively explore interventions less aggressively than UCB. In this talk, we present both theoretical and empirical benefits of this approach. Based on: https://arxiv.org/abs/2002.08405
Bio: Sanjay Shakkottai received his Ph.D. from the ECE Department at the University of Illinois at Urbana-Champaign in 2002. He is with The University of Texas at Austin, where he is currently the Temple Foundation Endowed Professor No. 3, and a Professor in the Department of Electrical and Computer Engineering. He received the NSF CAREER award in 2004, and was elected as an IEEE Fellow in 2014. His research interests lie at the intersection of algorithms for resource allocation, statistical learning and networks, with applications to wireless communication networks and online platforms. Zoom: https://arizona.zoom.us/j/183045568