BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1766
DTSTAMP:20260902T045724Z
SUMMARY:Late Interaction: What\, When and Why
DESCRIPTION:Speaker: Sabyasachi Basu (Microsoft Research\, Bangalore)\n\nAb
 stract: \nLate interaction models\, which represent documents using a set 
 of vectors (and are therefore also called multivector models)\, have emerg
 ed as compelling alternatives to single-vector embedding models across a v
 ariety of applications in modern ML systems. While their empirical advanta
 ges have been observed for almost half a decade\, they come with significa
 nt costs: their storage footprint is substantially larger\, and retrieval 
 latency is considerably higher\, making them less attractive for several a
 pplications. This raises a natural question: when do we actually need l
 ate interaction?\nIn recent years\, existing theoretical results have prov
 ided important insights into the expressive power of multivector represent
 ations\, but do not directly answer this question in the context of retrie
 val—their most natural application domain. In this talk\, we give an exp
 licit answer to when and why. We construct a family of retrieval proble
 ms for which late interaction models provably exhibit an exponential gap 
 in representation size over their single-vector counterparts. We then show
  how this construction naturally leads to a semantic retrieval benchmark e
 xhibiting the same qualitative separation in real-world models deployed in
  practice.\nOur result uses techniques from the theory of pattern matrices
 \, illustrating how ideas from TCS can explain an empirical phenomenon in 
 the world of ML and help direct future research. Finally\, while deploying
  late interaction in real-world systems raises substantial challenges and 
 has led to a rich body of engineering-focused work\, we will delegate the
  how to a later time.\nBio: Sabyasachi Basu is a postdoctoral researcher
  at Microsoft Research Bangalore\, where he is part of the DiskANN team\, 
 working on efficient retrieval and approximate nearest neighbor search. He
  earned his PhD in Computer Science from UC Santa Cruz in 2025\, under the
  supervision of C. Seshadhri. His work focuses on bridging theory and prac
 tice in computing research across a spectrum of classical algorithmic prob
 lems\, with a particular emphasis on large-scale graphs.\n
URL:https://www.tcs.tifr.res.in/web/events/1766
DTSTART;TZID=Asia/Kolkata:20260925T113000
DTEND;TZID=Asia/Kolkata:20260925T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR