BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:182@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20260225T110000
DTEND;TZID=Asia/Kolkata:20260225T120000
DTSTAMP:20260219T151019Z
URL:https://cds.iisc.ac.in/events/cds-kiac-seminar-cds-102-25th-february-t
 owards-reliable-llm-reasoning-coordinated-agents-variance-aware-evaluation
 -and-lean-inference/
SUMMARY:CDS-KIAC {Seminar} @ CDS: #102: 25th\, February "Towards Reliable L
 LM Reasoning: Coordinated Agents\, Variance-Aware Evaluation\, and Lean In
 ference
DESCRIPTION:We welcome you to CDS-KIAC talk on 25th February 2026 (Wednesda
 y). The details are as below:\n\n\n\nSpeaker : Dr. Akhil Arora\, Assistant
  professor at Aarhus University\nTitle : Towards Reliable LLM Reasoning: C
 oordinated Agents\, Variance-Aware Evaluation\, and Lean Inference\nDate a
 nd Time : February 25\, 2026: 11:00 AM\nVenue : The Seminar will be held o
 n HYBRID Mode\n# 102 CDS Seminar Hall /MICROSOFT TEAMS\nPlease click on th
 e following link to join the Seminar:\nMS Teams link\n\n\n\nABSTRACT:\nLar
 ge language models (LLMs) are increasingly deployed as reasoning engines\,
  yet their practical use remains constrained by three persistent challenge
 s: achieving high-quality reasoning at low cost\, measuring performance re
 liably\, and ensuring efficient\, reproducible deployment. In this talk\, 
 I will present a research agenda addressing these challenges through new m
 ethods\, benchmarks\, and systems for practical LLM reasoning. I begin wit
 h coordination as a pathway to efficiency. Fleet of Agents (FoA) introduce
 s a framework where swarms of lightweight LLM agents explore search spaces
  in parallel and are resampled through a genetic-style process. This desig
 n shows that orchestration can often matter more than sheer size\, enablin
 g smaller models to outperform larger ones while achieving superior cost-q
 uality trade-offs across diverse reasoning tasks. Next\, I turn to evaluat
 ion as a foundation for trust. ReasonBench exposes the limits of single-ru
 n reporting by systematically quantifying the run-to-run variability of LL
 M reasoning. Through variance-aware metrics\, it reveals the hidden instab
 ility and cost unpredictability of many reasoning strategies\, highlightin
 g reproducibility as a first-class requirement for reliable reasoning. Fin
 ally\, I focus on systems as enablers of sustainable deployment. I present
  CacheSaver\, the first modular client-side framework for high-level infer
 ence optimization. By introducing a namespace-aware caching mechanism\, Ca
 cheSaver reduces cost and carbon emissions while preserving statistical in
 tegrity\, making large-scale experimentation and deployment more affordabl
 e and sustainable without compromising reproducibility. Together\, these c
 ontributions chart a path toward LLM reasoning that is not only more power
 ful\, but also leaner\, more reliable\, and environmentally responsible.\n
 \nBIOGRAPHY:\nAkhil Arora is a Prof. of Computer Science at Aarhus Univers
 ity\, where he heads the CLAN for AI Research on Language and Networks (or
  “CLAN” for short). He is a fellow of the Copenhagen Center for Social
  Data Science (SODAS)\, an affiliate of ELLIS Unit Copenhagen\, the Pionee
 r Centre for AI (P1)\, and a formal collaborator of the Wikimedia Foundati
 on. Akhil’s research lies broadly in human-centered and trustworthy AI\,
  with a focus on improving human knowledge-seeking and promoting knowledge
  equity on the Web. His work blends methods from NLP\, AI\, Graph ML\, and
  computational social science\, including state-of-the-art work on structu
 red and multi-agent reasoning and systems for reliable\, cost-efficient de
 ployment of LLMs. Akhil received his PhD in Computer Science from EPFL (20
 24) in Switzerland\, his MS from IIT Kanpur (2013)\, and his undergraduate
  degree from NCU Gurgaon (2010). In days of yore\, he spent close to five 
 years in the industry working with the research labs of Xerox and American
  Express as a Research Scientist. His work on influence maximization has
  been recognized as the 8th most influential paper of SIGMOD 2017 by P
 aper Digest and received the 2018 ACM SIGMOD Most Reproducible Paper Award
 . He is a recipient of the prestigious EDIC Doctoral Fellowship\, an alumn
 us of the coveted Heidelberg Laureate Forum\, and a DAAD AINet fellow on h
 uman-centered AI. Akhil is a director of the P1-programs on Green AI and A
 I &amp\; Society.\n\nHost Faculty: Dr. Danish Pruthi\, CDS\n\n\n\nALL ARE 
 WELCOME
CATEGORIES:Events,Talks
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20250225T110000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR