[Selected Papers] [Full List] [Google Scholar] [DBLP] [arXiv]
Collaborators.
Throughout my journey, I have been incredibly fortunate to work alongside some of the brilliant minds in the field, whose wisdom has been instrumental in shaping my work and enriching my knowledge fundamentally: Hilal Asi, Viktor Bengs, Chiranjib Bhattacharyya, Avrim Blum, Christos Dimitrikakis, Simon Du, Maryzam Fazel, Yonathan Efroni, Vitaly Feldman Pierre Gaillard, Aditya Gopalan, Katja Hofmann Eyke Hüllermeier, Prateek Jain, Thomas Kleine Buening, Tomer Koren, Branislav Kveton, Akshay Krishnamurthy, Haipeng Luo, Shie Mannor, Yishay Mansour, Pasin Manurangsi, Nadav Merlis, Nagarajan Natarajan, Praneeth Netrapalli, Vianney Perchet, Lev Reyzin, Rob Schapire, Nati Srebro, Michal Valko, Arun Verma, Matthew Walter, Haifeng Xu, Brian Ziebert (in alphabetical order, sorry if not up to date!).
Preprints/ Working Drafts:
- Self-Consistency on a Budget [Arxiv version (soon!)]
Aniket Wadge, Branislav Kveton, Aadirupa Saha - Sample Efficient Policy Optimization with Different Feedback Modalities [Arxiv version (soon!)]
Aadirupa Saha, Pierre Gaillard - Instance-Optimal Dueling Bandits with Differential Privacy [Arxiv version (soon!)]
Aadirupa Saha, Pasin Manurangsi - Replicable Dueling Bandits [Arxiv version]
Rohan Deb, Aadirupa Saha, Udaya Ghai, Karan Singh, Arindam Banerjee - Preference Learning Beyond Confidence: Gradient-based RLHF with Adversarial Preferences [Arxiv version (soon!)]
Aadirupa Saha - Optimal Learning Rates with Costly Observations [Arxiv version (soon!)]
Lev Reyzin, Aadirupa Saha (alphabetical) - Double-Monster: Efficient Min-Max Strategy for Personalized Prediction under General Preferences.
Aadirupa Saha, Robert Schapire - Variance Adaptive Bandits under Cost Budget
Amith Bhat, Aadirupa Saha - Imitation Beyond Expectation via Second-order Stochastic Dominance
Syed M. Abbas, Danyal Saeed, Aadirupa Saha, Brian D Ziebart - Efficient Predictive Models without Compromising User Privacy [Arxiv version]
Aadirupa Saha, Hilal Asi - Learning to Allocate Resources with Censored Feedback [Arxiv version]
Giovanni Montanari, Côme Fiegel, Aadirupa Saha, Vianney Perchet - COBRA: Contextual Bandit Algorithm for Ensuring Truthful Strategic Agents [Arxiv Version]
Arun Verma, Indrajit Saha, Aadirupa Saha, Daniela Rus, Makoto Yokoo, Bryan Kian Hsiang Low - Best Arm Identification in Linear MNL-Bandits. [Arxiv Version]
Shubham Gupta, Aadirupa Saha, Sumeet Katariya
Full list of Publications:
2027
- Bandit PCA with Minimax Optimal Regret [Arxiv]
Moise Blanchard, Dmitry Ostrovsky, Aadirupa Saha (alphabetical)
In 38th ACM-SIAM Symposium on Discrete Algorithms (SODA), 2027
2026
- Expert Advice with Costly Observations
Lev Reyzin, Aadirupa Saha, Shuo Wu (alphabetical)
In Uncertainty in Artificial Intelligence, UAI 2026 - SOAR: Data-Source Adaptive Online Learning under Heteroscedastic Noisea [Arxiv version]
Amith Bhat, Haipeng Luo, Aadirupa Saha
International Conference on Machine Learning, ICML 2026 - LLM-as-a-Judge on a Budget [Arxiv version]
Aadirupa Saha, Branislav Kveton, Aniket Wagde
International Conference on Artificial Intelligence and Statistics, AIStats 2026 - Stochastically Dominant Preference Optimization: Policy Improvement for All [Arxiv version]
Ali Farajzadeh, Syed M Abbas, Aadirupa Saha, Brian D Ziebart
International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2026 (Extended Abstract)
2025
- Efficient and Near-Optimal Algorithm for General Contextual Dueling Bandits with Offline Regression Oracles [NeurIPS version]
Aadirupa Saha, Robert Schapire
In Neural Information Processing Systems, NeurIPS 2025 - Imitation Beyond Expectation Using Pluralistic Stochastic Dominance [NeurIPS version]
Ali Farajzadeh, Danyal Saeed, Syed M Abbas, Rushit N. Shah, Aadirupa Saha, Brian D Ziebart
In Neural Information Processing Systems, NeurIPS 2025 (*Spotlight*) - Source Adaptive Online Learning under Heteroscedastic Noise [OPT-ML version]
Amith Bhat, Aadirupa Saha, Thomas Kleine Buening, Haipeng Luo
In OPT for ML Workshop, Neural Information Processing Systems, NeurIPS 2025 - Efficient Algorithms for Combinatorial-Bandits with Monotonicity. [OPT-ML version]
Aniket Wadge, Aadirupa Saha
In OPT for ML Workshop, Neural Information Processing Systems, NeurIPS 2025 - HPO: Provably Faster Convergence Rates by Combining Offline Preferences with Online Exploration [Arxiv version] [Workshop version]
Avinandan Bose, Zhihan Xiong, Aadirupa Saha, Simon Shaolei Du, Maryam Fazel
In The Next Frontier in Reliable AI Workshop, International Conference on Learning Representations (ICLR), 2025 - Tracking the Best Expert Privately [Arxiv version]
Hilal Asi, Vinod Raman, Aadirupa Saha (alphabetical)
International Conference on Machine Learning, ICML 2025 - Dueling Convex Optimization for General Preferences: An Unified Framework for Optimal Convergence Rates. [Arxiv Version]
Aadirupa Saha, Tomer Koren, Yishay Mansour
International Conference on Machine Learning, ICML 2025 - Stop Relying on No-Choice and Do not Repeat the Moves: Optimal, Efficient and Practical Algorithms for Assortment Optimization [Arxiv version]
Aadirupa Saha, Pierre Gaillard
International Conference on Learning Representations (ICLR), 20251 min pitch!
If you've ever encountered MNL Assortment Optimization problem, the go-to approach is to offer the same set of products repeatedly until your customer is really annoyed and selects no item! In fact, it requires a "belief" that no selection is their most preferred choice :-( Oh no!
But why be so pessimistic? And why annoy your customers repeatedly offering the same items and hoping them to leave (i.e. they decide to choose none of the offered items!)? We got a new idea with no such issues. How? We simply found better concentration tricks! It was a long time wish to resolve this efficiently.
2024
- Strategic Linear Contextual Bandits. [Arxiv version]
Thomas Kleine Buening, Aadirupa Saha, Haifeng Xu, Christos Dimitrakakis
In Neural Information Processing Systems, NeurIPS 2024 - Dueling in the Dark: An Efficient and Optimal O(√T) Mirror Descent Approach for Competing against Adversarial Preferences. [Arxiv: Coming soon!]
Aadirupa Saha, Barry-John Theobald, Yonathan Efroni
In OPT for ML Workshop, Neural Information Processing Systems, NeurIPS 2024 - A Graph Theoretic Approach for Preference Learning with Feature Information. [Arxiv Version]
Aadirupa Saha, Arun Rajkumar
In Uncertainty in Artificial Intelligence, UAI 2024 (*Oral*) - Social Welfare for RecSys: Bandits Meet Mechanism Design to Combat Clickbait in Online Recommendation. [Arxiv version]
Thomas Kleine Buening, Aadirupa Saha, Haifeng Xu, Christos Dimitrakakis
International Conference on Learning Representations (ICLR), 2024 (*Spotlight*) - Only Pay for What Is Uncertain: Variance-Adaptive Thompson Sampling. [Arxiv version]
Aadirupa Saha, Branislav Kveton
International Conference on Learning Representations (ICLR), 20241 min pitch!
We lay the foundations for Bayesian multi-armed bandits with known and unknown heterogeneous reward variances with Thompson sampling. Our regret analysis shows improved performance with lower reward variances, implying faster learning in low-variance regimes. So why regret if you are already confident - Only Pay for What Is Uncertain! - Efficient Private Federated Non-Convex Optimization With Shuffled Model. [Workshop Version]
Lingxiao Wang, Xingyu Zhou, Kumar Kshitij Patel, Lawrence Tang, Aadirupa Saha
Privacy Regulation and Protection in ML Workshop, International Conference on Learning Representations (ICLR), 2024 - Think Before You Duel: Understanding Complexities of Preference Learning under Constrained Resources. [Arxiv version]
Rohan Deb, Aadirupa Saha
International Conference on Artificial Intelligence and Statistics, AIStats 2024 - On the Vulnerability of Fairness Constrained Learning to Malicious Noise. [Arxiv version]
Avrim Blum, Princewill Okoroafor, Aadirupa Saha, Kevin Stangl (alphabetical)
International Conference on Artificial Intelligence and Statistics, AIStats 2024 - Faster Convergence with MultiWay Preferences. [Arxiv version]
Aadirupa Saha, Vitaly Feldman, Tomer Koren, Yishay Mansour
International Conference on Artificial Intelligence and Statistics, AIStats 2024 - Dueling Optimization with a Monotone Adversary
Avrim Blum, Meghal Gupta, Gene Li, Naren Sarayu Manoj, Aadirupa Saha, Yuanyuan Yang
Algorithmic Learning Theory, ALT, 2024 (*Outstanding Paper Award*)
2023
- Eliciting User Preferences for Personalized Multi-Objective Decision Making through Comparative Feedback [Arxiv Version]
Han Shao, Lee Cohen, Avrim Blum, Yishay Mansour, Aadirupa Saha, Mathew Walter
In Neural Information Processing Systems, NeurIPS 2023 - Dueling Optimization with a Monotone Adversary. [Arxiv version]
Avrim Blum, Meghal Gupta, Gene Li, Naren Sarayu Manoj, Aadirupa Saha, Yuanyuan Yang
NeurIPS OPT+ML Workshop, NeurIPS, 2023 (Oral) - On the Vulnerability of Fairness Constrained Learning to Malicious Noise. [Arxiv version]
Avrim Blum, Princewill Okoroafor, Aadirupa Saha, Kevin Stangl
Algorithmic Fairness through the Lens of Time Workshop, NeurIPS, 2023 - Federated Online and Bandit Convex Optimization [Arxiv Version]
Kumar Kshitij Patel, Lingxiao Wang, Aadirupa Saha, Nati Srebro
In the International Conference on Machine Learning, ICML 2023 - Bandits Meet Mechanism Design to Combat Clickbait in Online Recommendation [Arxiv Version]
Thomas Kleine Buening, Aadirupa Saha, Haifeng Xu, Christos Dimitrakakis
Interactive Learning with Implicit Human Feedback Workshop, ICML 2023 - One Arrow, Two Kills: An Unified Framework for Achieving Optimal Regret Guarantees in Sleeping Bandits [Arxiv Version] [Talk]
Pierre Gaillard, Aadirupa Saha, Soham Dan
In International Conference on Artificial Intelligence and Statistics, AIStats 2023 - ANACONDA: Improved Dynamic Regret Algorithm for Adaptive Non-Stationary Dueling Bandits [Arxiv Version]
Thomas Kleine Buening, Aadirupa Saha
In International Conference on Artificial Intelligence and Statistics, AIStats 2023 - Dueling RL: Reinforcement Learning with Trajectory Preferences [Arxiv Version]
Aadirupa Saha*, Aldo Pacchiano*, Jonathan Lee (*Equal contribution)
In International Conference on Artificial Intelligence and Statistics, AIStats 2023
1 min pitch!
Sleeping Bandits are as interesting as they sound, but what is the right measure of Sleeping Regret? So many different notions of regrets were studied in the literature --- Sleeping External regret, Ordering regret, Policy regret --- but it is confusing to keep track of the implications of so many different notions, i.e. every combination of stochastic or adversarial losses and availability pairs.Can we unify them under a single measure? We found one in this work - Sleeping Internal Regret! One of our main contributions is unifying existing notions of regret in sleeping bandits and exploring their implications for each other.
Our proposed algorithm achieves sublinear Internal Regret, even when losses and availabilities are both adversarial, which is the hardest combination of sleeping setup! Further, our results show how a low internal regret leads to both low external regret and low policy regret - One arrow, Two Kills!
Our unified notion of sleeping regret also helps to invent a general notion of Sleeping Dueling Bandits that is stronger than the existing regret definitions used in the contemporary dueling bandits literature and overcomes the issue of repeated draws if needed. This is the first bound of this kind in the dueling literature with many potentials!
2022
- Distributed Online and Bandit Convex Optimization
Kumar Kshitij Patel, Aadirupa Saha, Lingxiao Wang, Nati Srebro
In OPT ML Workshop, Neural Information Processing Systems, NeurIPS 2022 - Versatile Dueling Bandits: Best-of-both-World Analyses for Online Learning from Preferences. [Arxiv Version]
Aadirupa Saha, Pierre Gaillard
In International Conference on Machine Learning, ICML 2022 - Optimal and Efficient Dynamic Regret Algorithms for Non-Stationary Dueling Bandits. [Arxiv Version]
Aadirupa Saha*, Shubham Gupta* (*Equal contribution)
In International Conference on Machine Learning, ICML 2022 - Stochastic Contextual Dueling Bandits under Linear Stochastic Transitivity Models. [Arxiv Version]
Viktor Bengs, Aadirupa Saha, Eyke Hüllermeier
In International Conference on Machine Learning, ICML 2022 - Efficient and Optimal Algorithms for Contextual Dueling Bandits under Realizability [Arxiv Version]
Aadirupa Saha, Akshay Krishnamurthy
In Algorithmic Learning Theory, ALT 2022 - Exploiting Correlation to Achieve Faster Learning Rates in Low-Rank Preference Bandits [Arxiv Version]
Aadirupa Saha*, Suprovat Ghoshal* (*Equal contribution)
In International Conference on Artificial Intelligence and Statistics, AIStats 2022
2021
- Dueling Bandits with Adversarial Sleeping [Arxiv Version]
Aadirupa Saha, Pierre Gaillard
In Neural Information Processing Systems, NeurIPS 2021 - Optimal Algorithms for Stochastic Contextual Dueling Bandits
Aadirupa Saha
In Neural Information Processing Systems, NeurIPS 2021 - Dueling Convex Optimization
Aadirupa Saha, Tomer Koren, Yishay Mansour
In International Conference on Machine Learning, ICML 2021 - Adversarial Dueling Bandits [Arxiv Version]
Aadirupa Saha, Tomer Koren, Yishay Mansour
In International Conference on Machine Learning, ICML 2021 - Optimal Regret Algorithm for Pseudo-1d Bandit Convex Optimization [Arxiv Version]
Aadirupa Saha, Nagarajan Natarajan, Praneeth Netrapalli, Prateek Jain
In International Conference on Machine Learning, ICML 2021 - Confidence-Budget Matching for Sequential Budgeted Learning [Arxiv Version]
Yonathan Efroni, Nadav Merlis, Aadirupa Saha, Shie Mannor
In International Conference on Machine Learning, ICML 2021 - Strategically Efficient Exploration in Competitive Multi-agent Reinforcement Learning [Arxiv Version]
Robert Loftin, Aadirupa Saha, Sam Devlin, Katja Hofmann
In Uncertainty in Artificial Intelligence, UAI 2021
2020
- From PAC to Instance-Optimal Sample Complexity in the Plackett-Luce Model [Arxiv Version]
Aadirupa Saha, Aditya Gopalan
In International Conference on Machine Learning, ICML 2020 - Improved Sleeping Bandits with Stochastic Action Sets and Adversarial Rewards [Arxiv Version]
Aadirupa Saha, Pierre Gaillard, Michal Valko
In International Conference on Machine Learning, ICML 2020 - Best-item Learning in Random Utility Models with Subset Choices [Arxiv Version]
Aadirupa Saha, Aditya Gopalan
In International Conference on Artificial Intelligence and Statistics, AIStats 2020 - Polytime Decomposition of Generalized Submodular Base Polytopes with Efficient Sampling
Aadirupa Saha
In Asian Conference on Machine Learning, ACML 2020
2019
- Combinatorial Bandits with Relative Feedback [Arxiv Version]
Aadirupa Saha, Aditya Gopalan
In Neural Information Processing Systems, NeurIPS 2019 - Be Greedy: How Chromatic Number meets Regret Minimization in Graph Bandits
Shreyas Seshadri*, Aadirupa Saha*, Chiranjib Bhattacharyya (*Equal Contribution)
In Uncertainty in Artificial Intelligence, UAI 2019 - Active Ranking with Subset-wise Preferences [Arxiv Version]
Aadirupa Saha, Aditya Gopalan
In International Conference on Artificial Intelligence and Statistics, AIStats 2019 - PAC Battling Bandits in the Plackett-Luce Model [Arxiv Version]
Aadirupa Saha, Aditya Gopalan
In Algorithmic Learning Theory, ALT 2019 - How Many Pairwise Preferences Do We Need to Rank A Graph Consistently? [Arxiv Version]
Aadirupa Saha, Rakesh Shivanna, Chiranjib Bhattacharyya
In AAAI Conference on Artificial Intelligence, AAAI 2019
2018
- Battle of Bandits
Aadirupa Saha, Aditya Gopalan
In Uncertainty in Artificial Intelligence, UAI 2018 - Online Learning for Structured Loss Spaces [Arxiv Version]
Siddharth Barman, Aditya Gopalan, Aadirupa Saha (alphabetical)
In AAAI Conference on Artificial Intelligence, AAAI 2018
2015
- Consistent Multiclass Algorithms for Complex Performance Measures
Harikrishna Narasimhan, Harish Ramaswamy, Aadirupa Saha, Shivani Agarwal
In International Conference on Machine Learning, ICML 2015
2014
- Learning Score Systems for Patient Mortality Prediction in Intensive Care Units via Orthogonal Matching Pursuit
Aadirupa Saha, Chandrahas Dewangan, Harikrishna Narasimhan, Sriram Sampath, Shivani Agarwal
In International Conference on Machine Learning and Applications, ICMLA 2014
2013
- Energy Saving Replay Attack Prevention in Clustered Wireless Sensor Networks
Amrita Ghosal, Aadirupa Saha, Sipra Das Bit
In Pacific-Asia Workshop on Intelligence and Security Informatics, PAISI 2013
2011
- Energy-Balancing and Lifetime Enhancement of Wireless Sensor Network with Archimedes Spiral
Subir Halder, Amrita Ghosal, Aadirupa Saha, Sipra DasBit
In International Conference on Ubiquitous Intelligence and Computing, ICUIC 2011
[Back to Top] [Selected Papers] [Full List] [Google Scholar] [DBLP] [arXiv]