BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Iowa State University CALS LAS Web Team//sites.iastate.edu//EN
BEGIN:VEVENT
UID:20260717T120000-4964-www.cs.iastate.edu
DTSTART:20260717T120000Z
SEQUENCE:0
TRANSP:OPAQUE
DTEND:20260717T130000Z
LOCATION:Zoom: https://iastate.zoom.us/j/98514548658
SUMMARY:MS Final Oral Exam: Dhawal Shah
CLASS:PUBLIC
DESCRIPTION:Low-Rank KV Cache Compression in LLMs: A Second-Order Output Me
 tric and Its LimitsThe key–value (KV) cache makes autoregressive decodin
 g in large language models fast\, but it grows with the length of the cont
 ext. At long context it becomes the main drain on a GPU’s memory\, both 
 in how much it has to store and in how fast it can be read. Compressing th
 e keys and values with a low-rank projection is a cheap\, training-free wa
 y to shrink it. The trouble is that existing methods help only a little at
  the compression ratios people actually use\, and the targets they optimiz
 e are mostly heuristics. This thesis builds a second-order account of atte
 ntion-aware low-rank KV compression and uses it to show both what these me
 thods can do and where they break down.\n\nMore information at: https://ww
 w.cs.iastate.edu/event/2026/ms-final-oral-exam-dhawal-shah\n\nZoom: https:
 //iastate.zoom.us/j/98514548658
DTSTAMP:20260809T125615Z
END:VEVENT
END:VCALENDAR