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:160@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20251202T100000
DTEND;TZID=Asia/Kolkata:20251202T110000
DTSTAMP:20251115T043802Z
URL:https://cds.iisc.ac.in/events/m-tech-research-thesis-colloquium-cds-im
 proving-out-of-distribution-robustness-in-long-tailed-medical-image-datase
 ts/
SUMMARY:M.Tech Research Thesis {Colloquium}: CDS: "Improving Out-of-Distrib
 ution Robustness in Long-Tailed Medical Image Datasets"
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nM.Tech Research 
 Thesis Colloquium\n\n\n\nSpeaker : Mr. Archakam Satvikam Anudeep\nS.R. Num
 ber : 06-18-01-10-22-23-1-23183\nTitle : "Improving Out-of-Distribution Ro
 bustness in Long-Tailed Medical Image Datasets"\nResearch Supervisor : Dr.
  Vaanathi Sundaresan\nDate &amp\; Time : December 02\, 2025\, 10.00 AM\nVe
 nue : # 102 CDS Seminar Hall\n\n\n\nABSTRACT\n\nDeep neural networks (DNNs
 ) have achieved remarkable success across medical imaging applications. Ho
 wever\, their clinical deployment remains limited due to vulnerability to 
 out-of-distribution (OOD) data\, i.e.\, inputs that differ from the traini
 ng distribution\, leading to overconfident yet incorrect predictions. This
  challenge is further compounded by class imbalance and domain shifts in r
 eal world clinical data. While long tailed OOD detection has been explored
 \, prior work mostly addresses natural images or narrowly targets medical 
 cases such as skin lesions\, with restricted evaluation on novel classes.\
 n\nTo address these gaps\, we propose a unified long-tailed OOD detection 
 framework for medical imaging\, designed and evaluated under clinically re
 alistic conditions. At its core lies a novel von Mises Fisher (vMF) based 
 classifier that learns nonlinear decision boundaries\, enabling more flexi
 ble and robust feature separation than conventional vMF formulations. To m
 itigate bias toward head classes in long-tailed distributions\, we employ 
 a multi-expert design where several margin-adjusted vMF classifiers specia
 lize across different regions of the label space. Their aggregated predict
 ions reduce imbalance and enhance generalization. In addition\, an outlier
  expert trained with an outlier exposure objective captures atypical featu
 re patterns and suppresses overconfidence on unseen data. Together\, these
  components form an ensemble that balances class representation and streng
 thens OOD robustness.\n\nFurthermore\, our work advances long-tailed OOD d
 etection in three key aspects. We train and evaluate our framework on thre
 e diverse medical imaging datasets—ISIC2019 (dermatology)\, NIHCXR (ches
 t radiography)\, and RFMiD (retinal imaging)—whereas most prior studies 
 focus solely on skin lesions. We further introduce a clinically reflective
  OOD evaluation spectrum that spans near OODs (e.g.\, corrupted images and
  unseen disease variants) to far OODs (cross-modality and non-medical doma
 ins)\, providing one of the first systematic assessments of distributional
  shifts encountered in clinical deployment. Within this realistic setup\, 
 we benchmark multiple state-of-the-art methods and demonstrate that our pr
 oposed approach achieves the best overall performance. Compared to state-o
 f-the-art methods\, our method reduces the average FPR95 by 10.95%\, 18.73
 %\, and 7.68% on ISIC2019\, NIHCXR\, and RFMiD respectively\, while outper
 forming the baseline by over 59% across all datasets. In addition to enhan
 ced OOD detection\, our framework attains the highest balanced accuracy fo
 r ID classification\, underscoring its effectiveness in learning robust de
 cision boundaries under long-tailed conditions.\n\n\n\nALL ARE WELCOME
CATEGORIES:Events,MTech Research Thesis Colloquium
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20241202T100000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR