Preprints and Publications

Link to my: Google scholar

Working Papers: 

• Data-driven Error Estimation: Upper Bounding Multiple Errors with No Technical Debt [Working Draft]

Sanath Kumar Krishnamurthy, Susan Athey, Emma Brunskill.

[Motivated by the goal of deploying sequential decision-making algorithms without needing complexity of class as an input, we develop data-driven methods to estimate an upper bound on an entire class of errors.]

• Selective Uncertainty Propagation in Offline RL [arXiv]

Sanath Kumar Krishnamurthy, Shrey Modi, Tanmay Gangwani, Sumeet Katariya, Branislav Kveton, Anshuka Rangi

[Developed offline RL algorithms that enable improved guarantees by exploiting the treatment effect size of action on next-state distributions.]

• Contextual Bandits in a Survey Experiment on Charitable Giving: Within-Experiment Outcomes versus Policy Learning [arXiv]

Susan Athey*, Undral Byambadalai*, Vitor Hadad*, Sanath Kumar Krishnamurthy*, Weiwen Leung*, Joseph Jay Williams*

[Took contextual bandits to the real world (charitable giving)!]

Accepted Papers:

• Towards Costless Model Selection in Contextual Bandits: A Bias-Variance Perspective [arXiv] 

Sanath Kumar Krishnamurthy*, Adrienne Propp*, Susan Athey.

AISTATS 2024.

[First approach to incorporate bias-variance trade-off into model selection for contextual bandits! Demonstrated the feasibility of costless model selection in contextual bandits under fairly benign conditions! Developed a novel misspecification test to evaluate prediction error bounds! The misspecification test enabled assumption-free guarantees in  [NeurIPS 2023].]

• Proportional Response: Contextual Bandits for Simple and Cumulative Regret Minimization [arXiv]

Sanath Kumar Krishnamurthy, Ruohan Zhan, Susan Athey, Emma Brunskill.

NeurIPS 2023.

[New form of uncertainty for contextual decision making! First general-purpose regression-based algorithm for simple regret minimization! Also, the first general-purpose algorithm that flexibly trades off simple and cumulative regret guarantees in contextual bandits!]

• Flexible and Efficient Contextual Bandits with Heterogeneous Treatment Effect Oracles [arXiv]

Aldo Gael Carranza*,  Sanath Kumar Krishnamurthy*, Susan Athey.

AISTATS 2023.

[Demonstrate the theoretical and empirical benefits of using heterogeneous treatment effect estimation in contextual bandits! Uncovered favorable properties of using R-learner for downstream decision-making tasks!]

• Adapting to Misspecification in Contextual Bandits with Offline Regression Oracles [arXiv]

 Sanath Kumar Krishnamurthy, Vitor Hadad, Susan Athey

ICML 2021.

[Quantified the bias-variance trade-off in contextual bandits! Developed algorithms that do not require the ``realizability" (zero bias) assumption! ]

• Tractable Contextual Bandits Beyond Realizability [arXiv]

Sanath Kumar Krishnamurthy, Vitor Hadad, Susan Athey

AISTATS 2021.

[Explored how additional uniformly sampled data could help with tackling sensitivity of existing contextual bandit algorithms to the``realizability" (zero bias) assumption. ]

• Approximation Algorithms for Maximin Fair Division [arXiv]

Siddharth Barman*, Sanath Kumar Krishnamurthy*

TEAC 2020 (Invited paper, a preliminary version appeared in EC 2017).

[Developed polynomial time algorithms to achieve approximate maximin allocations of indivisible goods. The first constant approximation factor guarantee for submodular valuations.]

• On the Proximity of Markets with Integral Equilibria [arXiv]

Siddharth Barman*, Sanath Kumar Krishnamurthy*

AAAI, 2019. 

[Developed polynomial time methods to search over a subset of Pareto optimal (PO) allocations.]

• Finding Fair and Efficient Allocations [arXiv]

       Siddharth Barman*, Sanath Kumar Krishnamurthy*, Rohit Vaish*

       EC, 2018.

[Developed the first pseudo-polynomial time algorithm for computing fair (EF1) and Pareto optimal (PO) allocations of indivisible goods among agents with additive valuations! This is technically fascinating because even checking if an allocation is PO is co-NP complete. Our techniques to search over such allocations were used in several subsequent papers. We also developed the first polynomial time algorithm to achieve the state-of-the-art approximation factor for maximizing Nash Social Welfare. ]

• Greedy Algorithms for Maximizing Nash Social Welfare [arXiv]

Siddharth Barman*, Sanath Kumar Krishnamurthy*, Rohit Vaish*

AAMAS, 2018 (nominated for best paper).

[Developed polynomial time algorithms to maximize Nash Social Welfare for binary valuations.]

• Groupwise Maximin Fair Allocation of Indivisible Goods [arXiv]

Siddharth Barman*, Arpita Biswas*, Sanath Kumar Krishnamurthy*, Y. Narahari*

AAAI, 2018.

[Introduced the notion of groupwise maximin fairness for fair division of indivisible goods.]

Other Manuscripts:

• Survey Bandits with Regret Guarantees [arXiv]

 Sanath Kumar Krishnamurthy, Susan Athey.

[* denotes equal contribution / alphabetical ordering]