Selected work

Representative Papers

Selection and Parallel Trends

Mar 17, 2022Social Science Research Network

This paper addresses how treatment-group selection threatens the parallel trends assumption in Difference-in-Differences (DiD) estimation—a critical yet under-characterized identification challenge. Method: We formally characterize the empirical content of this threat and derive necessary and sufficient conditions for parallel trends to hold under general selection mechanisms. We propose a “selection-driven bias decomposition framework” that systematically partitions DiD estimation bias into selection effects and time-varying heterogeneity effects, and develop operational benchmarking strategies—both with and without covariates—grounded in causal inference theory, selection modeling, and sensitivity analysis. Contribution/Results: Applied to the National Supported Work (NSW) experiment reanalysis, our approach quantifies and corrects selection bias, substantially improving the credibility of DiD estimates and the robustness of causal conclusions.

27 citations4 influentialRead paper

Projection Inference for set-identified SVARs

Apr 18, 2025

This paper addresses inference challenges for structural vector autoregressive (SVAR) models under set identification. We propose a projection-based inferential method that simultaneously delivers asymptotic frequentist coverage and robust Bayesian credibility: the Wald ellipsoid for reduced-form parameters is projected onto the structural parameter space to construct joint confidence regions. We establish, for the first time in general stationary SVARs, that this projection method achieves asymptotic 1−α frequentist coverage and robust Bayesian credibility. Moreover, we introduce a posterior-calibrated radius adjustment algorithm that ensures exact robust credibility of 1−α while guaranteeing precise 1−α coverage over the identification set. Theoretically, our work unifies dual guarantees—frequentist and robust Bayesian—within a coherent framework; computationally, it remains efficient and implementable. Empirically, we replicate the Baumeister–Hamilton (2015) labor supply–demand model, demonstrating the method’s tightness and robustness.

15 citations1 influentialRead paper

Estimating Effects of Long-Term Treatments

Jul 07, 2023ACM Conference on Economics and Computation

Accurately estimating the causal effects of long-term product interventions—such as UI redesigns or recommendation algorithm updates—in digital platforms remains challenging, as conventional short-term A/B tests fail to capture delayed and evolving impacts. To address this, we propose the first causal inference framework specifically designed for estimating long-term treatment effects. Our approach disentangles time-varying confounding from lagged treatment effects by explicitly modeling treatment duration as a key covariate. It integrates structural time-series modeling, doubly robust estimation, and dynamic causal graphs to enable counterfactual effect estimation without requiring costly long-duration experiments. Evaluated on real-world platform data, our method reduces long-term effect estimation error by 42% and achieves high-fidelity predictions across core metrics—including user retention rate and click-through rate—thereby significantly improving both the reliability and efficiency of long-horizon strategy evaluation.

7 citationsRead paper

Graph Neural Networks for Causal Inference Under Network Confounding

Nov 15, 2022

This paper addresses causal inference from a single large-scale network observation, confronting the high-dimensional network confounding challenge arising from simultaneous network interference and treatment selection endogeneity. To overcome the limitation of conventional methods—which rely on low-dimensional summaries of confounders—we propose two key innovations: (i) the first integration of graph neural networks (GNNs) into a causal inference framework, enabling end-to-end modeling of high-dimensional network confounding; and (ii) a “network-distance-decaying interference” assumption, imposing a low-dimensional structural constraint that ensures theoretical interpretability of shallow GNNs. Combining nonparametric causal modeling with simultaneous equations estimation, our approach delivers consistent estimation of the local average treatment effect under endogenous peer effects. Extensive experiments on synthetic and real-world network data demonstrate superior confounding control and estimation accuracy compared to existing methods.

4 citations1 influentialRead paper

Auditing Rust Crates Effectively

Feb 06, 2026

This work addresses the high cost of manual security auditing in Rust projects, which heavily rely on third-party crates. The authors propose Cargo Scan, a novel approach that integrates interactive side-effect analysis with call-graph construction, leveraging Rust’s type and module systems to precisely identify potentially dangerous code and trace contextual dependencies. The method enables composable, automated security assessments and enhances practicality through module-boundary tracking and persistent audit reports. Experimental evaluation demonstrates that when auditing the hyper crate and its dependencies, the median amount of code requiring manual review is reduced to 0.2% of the original total. Among the top 10,000 crates on crates.io, approximately 3,500 are automatically classified as safe, while the remaining crates exhibiting hazardous side effects are concentrated in only about 3%, necessitating targeted human inspection.

3 citationsRead paper

Recent publications

Latest Papers

MARC v1: An Open-Source Multi-Agent Framework for Clinical AI Reasoning and Coordination

Aug 13, 2026

This work addresses the lack of interpretability, traceability, and modularity in prompt-based approaches using monolithic large language models for clinical AI reasoning. To overcome these limitations, the authors propose a deterministic multi-agent collaborative architecture in which role-specific agents handle distinct tasks—namely information extraction, reasoning, answer generation, and evaluation—enabling explicit context propagation and intermediate result tracking. The framework incorporates a YAML-based configuration mechanism and a Decomposer module that automatically generates task-specific prompts, facilitating zero-code customization and deployment. The resulting system is open-source, model-agnostic, and fully traceable across all stages, eliminating the need for manual prompt engineering, supporting stage-level error attribution, and significantly enhancing usability and transparency for non-technical clinical users.

0 citationsRead paper

Agentic Auto-Research is Fuzz Testing

Aug 10, 2026

This work addresses the challenge of sparse feedback faced by autonomous scientific agents when exploring vast experimental spaces, where conventional generate-and-rank paradigms struggle to guide effective exploration. Drawing an analogy between autonomous research and gray-box fuzz testing, the study proposes a coverage-guided dense feedback mechanism that leverages intermediate experimental executions to obtain low-cost, high-frequency signals of cognitive progress, dynamically steering subsequent interventions. The approach integrates a feedback-driven search strategy with a safeguarded scientific validation pipeline to mitigate false discoveries arising from adaptive reuse. Experimental results demonstrate that the proposed mechanism effectively predicts genuine scientific progress, substantially improves the cost-efficiency of exploration, and reduces the rate of false discoveries.

0 citationsRead paper

Fixed-Threshold Peeling in Sublinear MPC: Round-Approximation Tradeoffs and Applications

Aug 10, 2026

This work studies efficient algorithms for fundamental graph problems—including density-based edge orientation, coloring, densest subgraph detection, and k-core decomposition—under the sublinear-memory MPC model using a fixed-threshold peeling strategy. The authors design novel multi-round parallel approximation algorithms that achieve a flexible trade-off between round complexity and approximation ratio. Their key contribution is the first algorithm to break the Θ(√log n) round-complexity barrier for the densest subgraph problem in this model, achieving a (4+ε)-approximation in Õ(log^{1/3} n) rounds. Additionally, they significantly improve the round-approximation trade-offs for k-core decomposition and edge orientation, advancing the state of the art in scalable graph processing under stringent memory constraints.

0 citationsRead paper

Position Bias in Ordinal Classification: A Systematic Evaluation

Aug 09, 2026

This study systematically investigates position bias in large language models (LLMs) for ordinal classification, where predictions are unduly influenced by semantically irrelevant factors such as label order and example placement in prompts, compromising system stability. Through controlled experiments across five datasets, the authors evaluate ten state-of-the-art LLMs, analyzing eight factors spanning prompt design, task formulation, and model architecture, while comparing pointwise, pairwise, and listwise reasoning paradigms alongside multiple debiasing strategies. Findings reveal that all models exhibit sensitivity to three types of position bias; reducing label cardinality concurrently improves both accuracy and stability; listwise methods generally perform best but show inconsistent generalization; and existing debiasing techniques lack universal efficacy. Crucially, predictive performance and stability often diverge, with robustness depending on the entire system configuration rather than the model alone.

0 citationsRead paper

Integrating spectral and morphological plant features with decision-tree models for early-season cotton biomass and nitrogen status estimation from multi-year UAV data

Aug 07, 2026

This study addresses the need for precise estimation of biomass and nitrogen status during early cotton growth to support variable-rate fertilization. By integrating spectral indices derived from UAV-based multispectral imagery with morphological traits—such as plant height and canopy cover—and employing machine learning models including Random Forest Regression (RFR) and Extreme Gradient Boosting (XGB), the research predicts dry matter weight, plant nitrogen uptake, and nitrogen concentration. Innovatively leveraging multi-year field trials with spatiotemporally consistent spectral and morphological features, the study evaluates model generalizability through leave-one-year-out cross-validation. Results demonstrate that RFR and XGB achieve superior performance, with mean absolute percentage error (MAPE) for nitrogen concentration estimation below 8%. The derived Nitrogen Nutrition Index effectively discriminates among multiple levels of nitrogen stress.

0 citationsRead paper