PhD Dissertation Proposal: Hrishikesh Kulkarni

Title: “Set, Ranked and Generative IR: Addressing IR challenges in low-resource, low-latency and fragmented-context scenarios”

Information Retrieval (IR) is searching, locating and delivering most relevant information based on user information needs. This works in various formats and uses different information retrieval paradigms. Popular ways of IR include Set IR, Ranked IR and Generative IR. In this dissertation proposal, I argue that there are certain limitations in the current state-of-the-art social media set retrieval in low-resource settings, ranked retrieval in low-latency settings and generative retrieval based question-answering in fragmented-context settings. To overcome these limitations, I further propose approaches for addressing discrepancies in low-resource scenarios, establishing new effectiveness-efficiency pareto-frontier in low-latency settings and generating comprehensive and more relevant answers using generative models in fragmented-context settings.

Committee members:
Ophir Frieder (co-adviser)
Nazli Goharian (co-adviser)
Jeremy Fineman
Sean Macavaney (University of Glasgow)