Anurag Singh

I am a third-year PhD student at Rational Intelligence Group in CISPA Helmholtz Center for Information Security, Saarbrücken, Germany, advised by Dr. Krikamol Muandet. I am broadly interested in Trustworthy Machine Learning and Imprecise Probability.
Previously, I did my Master's in Computer Science at the TU Munich, where I was advised by Prof. Matthias Neissner and Prof. Debarghya Ghoshdastidar. I did my BEng. from Netaji Subhas Institute of Technology, University of Delhi.

CV / Google Scholar / Blog / LinkedIn

Recent News

[20/03/2026] New Pre-Print on Incentive Aware AI Regulations arXiv!
[20/03/2026] Rajeev's latest work on AI Regulatory Markets got accepted at ICLR AIMS 2026. Congratulations everyone!

Selected Publications

Incentive Aware AI Regulations: A Credal Characterisation Anurag Singh, Julian Rodemann, Rajeev Verma, Siu Lun Chau, and Krikamol Muandet PAIG Workshop at EurIPS 2025 (Oral Presentation) [Code] We study provide conceptual insights into the design of AI regulations that are robust to uncertainty and strategic behavior.
Truthful Elicitation of Imprecise Forecasts Anurag Singh, Siu Lun Chau, and Krikamol Muandet UAI 2025 (Oral Presentation 3% acceptance rate) [Code] We devise a novel method to elicit credal sets truthfully, overcoming prior impossibility results.
Domain Generalisation via Imprecise Learning Anurag Singh, Siu Lun Chau, Shaheen Bouabid, Krikamol Muandet ICML 2024 (Spotlight Presentation 3.5% acceptance rate) [Code] We contend that domain generalisation encompasses both statistical learning and decision-making. Learners are thus compelled to make normative judgments, leading to misalignment amidst institutional separation from model operators.

Other Publications (Click to expand!)

Bayesian Optimization for Building Social-Influence-Free Consensus Masaki Adachi, Siu Lun Chau, Wenjie Xu, Anurag Singh, Michael A Osborne, and Krikamol Muandet ArXiv 2025 We introduce Social Bayesian Optimization (SBO), a vote-efficient algorithm for consensus-building in collective decision-making.
Robust Feature Inference: A Test-time Defense Strategy using Spectral Projections Anurag Singh, Mahalakshmi Sabanayagam, Krikamol Muandet, Debarghya Ghoshdastidar TMLR 2024 Robust Features are learned first during training. We theoretically show this and use this insight to develop a test-time defense strategy.
Improving Semi-Supervised Domain Adaptation Using Effective Target Selection and Semantics Anurag Singh, Naren Doraiswamy, Sawa Takamuku, Megh Bhalerao, Titir Dutta Soma Biswas, Aditya Chepuri CVPR Workshop Learning with Limited and Imperfect Data, 2021 Talk
Adaptive Margin Diversity Regularizer for handling Data Imbalance in Zero-Shot SBIR Titir Dutta, Anurag Singh, Soma Biswas ECCV, 2020 (Spotlight Presentation 5% acceptance rate) We analyze the effect of class-imbalance on generalization to unseen classes for the ZS-SBIR. The work then proposes a novel regularizer termed AMDReg, which can seamlessly be used with several ZS-SBIR methods to improve their performance.
StyleGuide: Zero-Shot Sketch-based Image Retrieval Using Style-Guided Image Generation Titir Dutta, Anurag Singh, Soma Biswas IEEE Transactions on Multimedia, 2020 A style-guided image generation during retrieval to eliminate the effect of domain difference and intra-class variations.
Image Corpus Representative Summarization Anurag Singh, Lakshay Virmani, AV Subramanyam IEEE International Conference on Multimedia Big Data, 2019 (Honourable Mention Award, Best Paper Nomination) Poster This work is part of my undergraduate thesis which focused on developing end to end deep learning architecture for problem of Image Collection Summarization. We introduced a task-specific loss to generate the summary related to a given task. We also proposed analysis of goodness of summary by training a classifier on it and comparing its performance with entire dataset.