Statistics and Data Science Seminar - Sahil Loomba
— 5:00pm
Location:
In Person
-
Steinberg Auditorium, Baker Hall A53
Speaker:
SAHIL LOOMBA
,
Chapman–Schmidt AI in Science Fellow, Imperial College London
https://sloomba.github.io/
The behavior of complex social systems arises jointly from their structure — how individuals are connected — and their mechanism — how individuals influence one another. In many settings, these components are only partially observed: structures are measured incompletely, mechanisms may be misspecified, and behaviors are observed only under experimentally realized treatments, as in the fundamental problem of causal inference.
In this talk, I will examine how partial observability creates foundational challenges for both statistical inference and the design of interventions in social systems. From a statistical perspective, I will discuss how asymptotic arguments can be used to analytically estimate global connectivity properties of social networks from incomplete structural observations, like egocentric samples, enabling inferences about shortest paths in population-level systems.
From a causal and decision-theoretic perspective, I will argue that partial behavioral observability in social networks requires a rethinking of canonical causal estimands. I will show that there is a fundamental tension between individual-level causal contrasts and policy-relevant social objectives, and that this tension can only be resolved by focusing on expected average outcomes over a space of treatment policies, yielding a unifying framework for causal inference and decision-making in networked social interventions.
This perspective naturally leads to new challenges in off-policy causal estimation: I will discuss how one can evaluate the effects of networked interventions different from those experimentally implemented, and how these ideas connect to testing and learning underlying behavioral mechanisms. I will conclude by discussing how these insights inform the design of optimal experiments and effective interventions in social networks under partial observability.
—
Sahil Loomba is a Chapman–Schmidt AI in Science Fellow at Imperial College London. He was previously a Schmidt Science Fellow at the Institute for Data, Systems, and Society in the Schwarzman College of Computing at MIT, and an EPSRC Doctoral Prize Fellow at Imperial, where he also earned his PhD in Mathematics. His doctoral thesis on sparse and partially observed large-scale networks was awarded the Yael Naim Dowker Prize. His research lies at the intersection of statistics, applied probability, and causal inference, focusing on understanding connectivity, behavior, and interventions in large-scale social systems, with applications in public health.
For More Information:
jpaschke@andrew.cmu.edu