About

Kyra Gan is an Assistant Professor in the School of Operations Research and Information Engineering and Cornell Tech at Cornell University. Her research interests include adaptive and online algorithm design in personalized treatment, such as micro-randomized trials and N-of-1 trials, particularly under constraint settings. She also focuses on computerized and automated inference methods, including targeted learning with Reproducing Kernel Hilbert Spaces (RKHS), as well as robust causal discovery in medical data and issues of fairness in organ transplants. Her broader goal is to bridge the gap between research and practice in healthcare. Prior to her current position, she was a postdoctoral fellow at the Statistical Reinforcement Lab at Harvard University. She earned her Ph.D. in Operations Research from Carnegie Mellon University at the Tepper School of Business in 2022 and holds B.A.s in Mathematics and Economics from Smith College, obtained in 2017. Her work has been recognized with awards such as the 2021 Pierskalla Best Paper Award and the 2021 CHOW Best Student Paper Award in Operations Research and Management Science.

Research topics

• Economic growth
• Medicine
• Demography
• Internal medicine

Selected publications

• Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation

  Open MIND · 2026-02-12

  preprint

  Estimating longitudinal treatment effects is essential for sequential decision-making but is challenging due to treatment-confounder feedback. While Iterative Conditional Expectation (ICE) G-computation offers a principled approach, its recursive structure suffers from error propagation, corrupting the learned outcome regression models. We propose D3-Net, a framework that mitigates error propagation in ICE training and then applies a robust final correction. First, to interrupt error propagation during learning, we train the ICE sequence using Sequential Doubly Robust (SDR) pseudo-outcomes, which provide bias-corrected targets for each regression. Second, we employ a multi-task Transformer with a covariate simulator head for auxiliary supervision, regularizing representations against corruption by noisy pseudo-outcomes, and a target network to stabilize training dynamics. For the final estimate, we discard the SDR correction and instead use the uncorrected nuisance models to perform Longitudinal Targeted Minimum Loss-Based Estimation (LTMLE) on the original outcomes. This second-stage, targeted debiasing ensures robustness and optimal finite-sample properties. Comprehensive experiments demonstrate that our model, D3-Net, robustly reduces bias and variance across different horizons, counterfactuals, and time-varying confoundings, compared to existing state-of-the-art ICE-based estimators.

  DOI

• Integrating Causal DAGs in Deep RL: Activating Minimal Markovian States with Multi-Order Exposure

  ArXiv.org · 2026-05-08

  articleOpen accessSenior author

  Online reinforcement learning (RL) relies on the Markov property for guaranteed performance, but real-world applications often lack well-defined states given raw observed variables. While causal RL has attracted growing interest, existing work typically assumes Markovian states are provided and focuses on using causality to accelerate learning, leaving a fundamental gap: \emph{given a longitudinal causal graph over observed variables, how does one construct MDP states that provably satisfy the Markov property?} We address this by providing a procedure that constructs a provably minimal state representation. In deep RL, we observe that the minimal representation alone empirically fails to improve performance, indicating that neural networks cannot directly exploit Markovian minimality. To address this, we propose \textbf{MOSE} (Multi-Order State Exposure), which feeds multi-order historical state constructions into the same $Q$-function. MOSE consistently outperforms both the minimal state construction and single-window policies on common benchmarks and synthetic datasets. Including the minimal representation alongside MOSE can further improve performance. Our results establish a core principle for causal deep RL: minimal sufficiency is not enough, and \emph{controlled redundancy} is necessary to unlock the benefit of causal state information.

  PublisherOA PDF

• Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces

  arXiv (Cornell University) · 2026-05-12

  preprintOpen accessSenior author

  A critical step for reliable large language models (LLMs) use in healthcare is to attribute predictions to their training data, akin to a medical case study. This requires token-level precision: pinpointing not just which training examples influence a decision, but which tokens within them are responsible. While influence functions offer a principled framework for this, prior work is restricted to autoregressive settings and relies on an implicit assumption of token independence, rendering their identified influences unreliable. We introduce a flexible framework that infers token-level influence through a latent mediation approach for general prediction tasks. Our method attaches sparse autoencoders to any layer of a pretrained LLM to learn a basis of approximately independent latent features. Unlike prior methods where influence decomposes additively across tokens, influence computed over latent features is inherently non-decomposable. To address this, we introduce a novel method using Jacobian-vector products. Token-level influence is obtained by propagating latent attributions back to the input space via token activation patterns. We scale our approach using efficient inverse-Hessian approximations. Experiments on medical benchmarks show our approach identifies sparse, interpretable sets of tokens that jointly influence predictions. Our framework enhances trust and enables model auditing, generalizing to high-stakes domain requiring transparent and accountable decisions.

  PublisherDOI

• Integrating Causal DAGs in Deep RL: Activating Minimal Markovian States with Multi-Order Exposure

  arXiv (Cornell University) · 2026-05-08

  preprintOpen accessSenior author

  Online reinforcement learning (RL) relies on the Markov property for guaranteed performance, but real-world applications often lack well-defined states given raw observed variables. While causal RL has attracted growing interest, existing work typically assumes Markovian states are provided and focuses on using causality to accelerate learning, leaving a fundamental gap: \emph{given a longitudinal causal graph over observed variables, how does one construct MDP states that provably satisfy the Markov property?} We address this by providing a procedure that constructs a provably minimal state representation. In deep RL, we observe that the minimal representation alone empirically fails to improve performance, indicating that neural networks cannot directly exploit Markovian minimality. To address this, we propose \textbf{MOSE} (Multi-Order State Exposure), which feeds multi-order historical state constructions into the same $Q$-function. MOSE consistently outperforms both the minimal state construction and single-window policies on common benchmarks and synthetic datasets. Including the minimal representation alongside MOSE can further improve performance. Our results establish a core principle for causal deep RL: minimal sufficiency is not enough, and \emph{controlled redundancy} is necessary to unlock the benefit of causal state information.

  PublisherDOI

• Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces

  ArXiv.org · 2026-05-12

  articleOpen accessSenior author

  A critical step for reliable large language models (LLMs) use in healthcare is to attribute predictions to their training data, akin to a medical case study. This requires token-level precision: pinpointing not just which training examples influence a decision, but which tokens within them are responsible. While influence functions offer a principled framework for this, prior work is restricted to autoregressive settings and relies on an implicit assumption of token independence, rendering their identified influences unreliable. We introduce a flexible framework that infers token-level influence through a latent mediation approach for general prediction tasks. Our method attaches sparse autoencoders to any layer of a pretrained LLM to learn a basis of approximately independent latent features. Unlike prior methods where influence decomposes additively across tokens, influence computed over latent features is inherently non-decomposable. To address this, we introduce a novel method using Jacobian-vector products. Token-level influence is obtained by propagating latent attributions back to the input space via token activation patterns. We scale our approach using efficient inverse-Hessian approximations. Experiments on medical benchmarks show our approach identifies sparse, interpretable sets of tokens that jointly influence predictions. Our framework enhances trust and enables model auditing, generalizing to high-stakes domain requiring transparent and accountable decisions.

  PublisherOA PDF

• Smooth Multi-Policy Causal Effect Estimation in Longitudinal Settings

  arXiv (Cornell University) · 2026-05-14

  preprintOpen access

  Comparative evaluation of multiple dynamic treatment policies is essential for healthcare and policy decisions, yet conventional longitudinal causal inference methods estimate each in isolation, preventing information sharing across counterfactuals. We demonstrate that this separate estimation paradigm induces a structurally uncontrolled second-order bias, inflating finite-sample variance even after standard debiasing with longitudinal targeted maximum likelihood estimation(LTMLE). To address this, we propose a policy-aware reparameterization of Iterative Conditional Expectation (ICE) Q-functions that enables joint estimation through shared representations. We implement this approach in the Policy-Encoded Q Network (PEQ-Net), an architecture centered on a shared policy encoder. The encoder is trained using kernel mean embeddings, ensuring that the learned representation space reflects population-level policy dissimilarities. After applying an LTMLE correction step, we prove this design imposes a structural constraint on the second-order remainder, thereby stabilizing finite-sample variance. Experiments on semi-synthetic datasets demonstrate that PEQ-Net consistently outperforms existing ICE-based methods, achieving substantial reductions in root-mean-square error, particularly when evaluating closely related policies.

  PublisherDOI

• Deep Doubly Debiased Longitudinal Effect Estimation with ICE G-Computation

  ArXiv.org · 2026-02-12

  articleOpen access

  Estimating longitudinal treatment effects is essential for sequential decision-making but is challenging due to treatment-confounder feedback. While Iterative Conditional Expectation (ICE) G-computation offers a principled approach, its recursive structure suffers from error propagation, corrupting the learned outcome regression models. We propose D3-Net, a framework that mitigates error propagation in ICE training and then applies a robust final correction. First, to interrupt error propagation during learning, we train the ICE sequence using Sequential Doubly Robust (SDR) pseudo-outcomes, which provide bias-corrected targets for each regression. Second, we employ a multi-task Transformer with a covariate simulator head for auxiliary supervision, regularizing representations against corruption by noisy pseudo-outcomes, and a target network to stabilize training dynamics. For the final estimate, we discard the SDR correction and instead use the uncorrected nuisance models to perform Longitudinal Targeted Minimum Loss-Based Estimation (LTMLE) on the original outcomes. This second-stage, targeted debiasing ensures robustness and optimal finite-sample properties. Comprehensive experiments demonstrate that our model, D3-Net, robustly reduces bias and variance across different horizons, counterfactuals, and time-varying confoundings, compared to existing state-of-the-art ICE-based estimators.

  PublisherOA PDF

• Smooth Multi-Policy Causal Effect Estimation in Longitudinal Settings

  ArXiv.org · 2026-05-14

  articleOpen access

  Comparative evaluation of multiple dynamic treatment policies is essential for healthcare and policy decisions, yet conventional longitudinal causal inference methods estimate each in isolation, preventing information sharing across counterfactuals. We demonstrate that this separate estimation paradigm induces a structurally uncontrolled second-order bias, inflating finite-sample variance even after standard debiasing with longitudinal targeted maximum likelihood estimation(LTMLE). To address this, we propose a policy-aware reparameterization of Iterative Conditional Expectation (ICE) Q-functions that enables joint estimation through shared representations. We implement this approach in the Policy-Encoded Q Network (PEQ-Net), an architecture centered on a shared policy encoder. The encoder is trained using kernel mean embeddings, ensuring that the learned representation space reflects population-level policy dissimilarities. After applying an LTMLE correction step, we prove this design imposes a structural constraint on the second-order remainder, thereby stabilizing finite-sample variance. Experiments on semi-synthetic datasets demonstrate that PEQ-Net consistently outperforms existing ICE-based methods, achieving substantial reductions in root-mean-square error, particularly when evaluating closely related policies.

  PublisherOA PDF

• Fast Non-Episodic Finite-Horizon RL with K-Step Lookahead Thresholding

  arXiv (Cornell University) · 2026-01-31

  articleOpen accessSenior author

  Online reinforcement learning in non-episodic, finite-horizon MDPs remains underexplored and is challenged by the need to estimate returns to a fixed terminal time. Existing infinite-horizon methods, which often rely on discounted contraction, do not naturally account for this fixed-horizon structure. We introduce a modified Q-function: rather than targeting the full-horizon, we learn a K-step lookahead Q-function that truncates planning to the next K steps. To further improve sample efficiency, we introduce a thresholding mechanism: actions are selected only when their estimated K-step lookahead value exceeds a time-varying threshold. We provide an efficient tabular learning algorithm for this novel objective, proving it achieves fast finite-sample convergence: it achieves minimax optimal constant regret for $K=1$ and $\mathcal{O}(\max((K-1),C_{K-1})\sqrt{SAT\log(T)})$ regret for any $K \geq 2$. We numerically evaluate the performance of our algorithm under the objective of maximizing reward. Our implementation adaptively increases K over time, balancing lookahead depth against estimation variance. Empirical results demonstrate superior cumulative rewards over state-of-the-art tabular RL methods across synthetic MDPs and RL environments: JumpRiverswim, FrozenLake and AnyTrading.

  PublisherOA PDF

• Fast Non-Episodic Finite-Horizon RL with K-Step Lookahead Thresholding

  Open MIND · 2026-01-31

  preprintSenior author

  Online reinforcement learning in non-episodic, finite-horizon MDPs remains underexplored and is challenged by the need to estimate returns to a fixed terminal time. Existing infinite-horizon methods, which often rely on discounted contraction, do not naturally account for this fixed-horizon structure. We introduce a modified Q-function: rather than targeting the full-horizon, we learn a K-step lookahead Q-function that truncates planning to the next K steps. To further improve sample efficiency, we introduce a thresholding mechanism: actions are selected only when their estimated K-step lookahead value exceeds a time-varying threshold. We provide an efficient tabular learning algorithm for this novel objective, proving it achieves fast finite-sample convergence: it achieves minimax optimal constant regret for $K=1$ and $\mathcal{O}(\max((K-1),C_{K-1})\sqrt{SAT\log(T)})$ regret for any $K \geq 2$. We numerically evaluate the performance of our algorithm under the objective of maximizing reward. Our implementation adaptively increases K over time, balancing lookahead depth against estimation variance. Empirical results demonstrate superior cumulative rewards over state-of-the-art tabular RL methods across synthetic MDPs and RL environments: JumpRiverswim, FrozenLake and AnyTrading.

  DOI

Frequent coauthors

• Sridhar Tayur

  5 shared

• Andrew A. Li

  Carnegie Mellon University

  4 shared

• Esmaeil Keyvanshokooh

  4 shared

• Sridhar Tayur

  4 shared

• Zachary C. Lipton

  4 shared

• Andrew Li

  Carnegie Mellon University

  4 shared

• Susan A. Murphy

  3 shared

• Su Jia

  Cornell University

  3 shared

Education

• PhD, Operations Research

  Carnegie Mellon University

  2022

• BA, Mathematics

  Smith College

  2017

Awards & honors

• INFORMS Pierskalla Best Paper Award (2021)
• CHOW Best Student Paper Award in the Category of Operations…
• Tata Consultancy Services Fellowship (2020)