BEGIN:VCALENDAR
PRODID:-//github.com/ical-org/ical.net//NONSGML ical.net 4.0//EN
VERSION:2.0
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:STANDARD
DTSTART:20241103T020000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20250309T020000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DESCRIPTION:Title: HypRAG: Hyperbolic Dense Retrieval for Retrieval Augmen
 ted Generation Abstract: Embedding geometry fundamentally affects retriev
 al quality\, yet dense retrievers for retrieval-augmented generation (RAG
 ) remain confined to Euclidean space. Natural language has hierarchical s
 tructure from broad topics to specific entities that Euclidean embeddings
  fail to preserve\, causing semantically distant documents to appear spur
 iously similar and increasing hallucination risk. We introduce hyperbolic
  dense retrieval with two model variants in the Lorentz model: HyTE-FH\, 
 a fully hyperbolic transformer\, and HyTE-H\, a hybrid architecture proje
 cting pre-trained Euclidean embeddings into hyperbolic space. To prevent 
 representational collapse during sequence aggregation\, we introduce the 
 Outward Einstein Midpoint\, a geometry-aware pooling operator that provab
 ly preserves hierarchical structure. On MTEB\, HyTE-FH outperforms equiva
 lent Euclidean baselines. On RAGBench\, HyTE-H achieves up to 29% gains o
 ver Euclidean baselines in context and answer relevance using substantial
 ly smaller models than current state-of-the-art retrievers. Hyperbolic re
 presentations encode document specificity through norm-based separation\,
  with over 20% radial increase from general to specific concepts\, a prop
 erty absent in Euclidean embeddings. Hiren Madhu a second year PhD studen
 t at Yale CS \, advised by Professor Smita Krishnaswamy and Professor Rex
  Ying . Earlier\, he was an ECE pre-doctoral research fellow in Electrica
 l Communication Engineering Department at the Indian Institute of Science
 \, Bengaluru. He worked with Sundeep Prabhakar Chepuri on unsupervised re
 presentation learning for simplicial complexes. He began his research jou
 rney at LDRP Institute of Technology and Research (BE in Computer Enginee
 ring) under the guidance of Sandip Modha and Thomas Mandl\, where he focu
 sed on the detection of hate speech in public conversational threads on s
 ocial media. His primary research interest is encoding the geometric indu
 ctive biases into foundation models. Previously\, he worked on simplicial
  representation learning without labels.\n\nSpeaker:\nHiren Madhu\n\nAdmi
 ssion:\nFree\n\nDetails URL:\nhttps://medicine.yale.edu/event/nlpllm-inte
 rest-group-25/\n
DTEND;TZID=America/New_York:20260302T170000
DTSTAMP:20260502T055826Z
DTSTART;TZID=America/New_York:20260302T160000
LOCATION:Zoom: Passcode Required\, URL: https://yale.zoom.us/j/93599941969
SEQUENCE:0
STATUS:Confirmed
SUMMARY:NLP/LLM Interest Group
UID:446de8c6-3d1a-4a98-96c8-e2fa527ca2c8
END:VEVENT
END:VCALENDAR