VisDom Lab, Data Science and Engineering, IISER Bhopal
rajeevias95 [at] gmail [dot] com rajeev22 [at] iiserb [dot] ac [dot] in
@phdthesis{dwivedi_phd,
author = {Rajeev Ranjan Dwivedi},
title = {Work in progress},
school = {IISER Bhopal},
advisor = {Vinod K. Kurmi},
year = {2022--present},
note = {Funded by the TCS Research Scholar Program.
Accepts coffee as input.}
}
Abstract—I am a PhD student in the Data Science and Engineering Department at IISER Bhopal, where I work with Dr. Vinod Kr. Kurmi in VisDom Lab. My PhD is funded by the TCS Research Scholar Program. I study why vision models learn shortcuts, such as a waterbird recognised by its water rather than its beak, and how to make them work for the groups those shortcuts leave behind. Recent work appeared at CVPR 2026 and ICLR 2026. Most of it is some version of Eq. (1): train a model so the group it serves worst is still served well.
I came to machine learning from statistics: an Integrated M.Sc. at Central University of Rajasthan and a thesis on heavy-tailed distributions. Dr. Sarika Jain at NIT Kurukshetra introduced me to NLP and the Semantic Web. This summer I was a PhD Tech Intern at Adobe Research, and before that I interned at Sony AI.
Index Terms—fairness in machine learning, spurious correlations, domain adaptation, uncertainty estimation, knowledge graphs.
Fig. 2. One latent task, ten prompt cells, one frozen model.
WSDM 2027 Findings · Telling an LLM to doubt itself, on top of telling it to check, moves accuracy by −1.4 to +2.6 points, in either direction.
Fig. 3. One AGS training step.
ICLR 2026 · A plug-in regularizer that damps shortcut neurons during training, with no group labels.
Fig. 4. PLD-Debias on a frozen backbone.
CVPR 2026 · Debiasing frozen foundation models with pseudo-bias labels and no access to the backbone.
Fig. 5. Retrieve cheaply, then re-rank with a token budget.
ICDM 2026 · Recovers 98.4% of full-token retrieval quality on a heavily compressed index.
Harnesses for large language models, focused on expert systems for high-stakes question answering.
May 2026 – Jul 2026LLM-based explainability over knowledge graphs for scientific discovery, and path-based graph traversal for transparent multi-hop reasoning.
Dec 2023 – Mar 2024Advisor: Dr. Vinod Kr. Kurmi. TCS Research Fellow.
Jul 2022 – presentMinor in Economics, Computer Science and Mathematics. Thesis with Dr. Deepesh Bhati: the Skew Laplace Slash Beta distribution, a new member of the slash family that is asymmetric, heavy-tailed and robust to outliers.
Jul 2016 – Aug 2021ICLR 2026 Financial Assistance, $1,600
2026WACV 2026 Doctoral Consortium travel grant, $1,500
2026Seed grant, IICE, IISER Bhopal, for NirVahan
2025TCS Research Scholar Program fellowship, 2024–2028
2024Prof. Ram Kumar Fellowship, GAME-ARTS, IISc Bangalore
2024ACM India Anveshan Setu Fellowship, a PhD visitor program; hosted by Dr. Vineeth N Balasubramanian, IIT Hyderabad
2023I read books on business and economics, and watch movies sometimes. I also enjoy cooking, and I usually cannot stop talking about South Indian cuisine. At university I started the campus magazine Astitva and was its chief editor for two years.
If you are nearby, let’s get coffee. We can talk about tech, data, startups, or anything else we both find interesting. Pick a time here.
