Selected work

Representative Papers

VoCa: Designing Speech-Canvas Interaction for Voice-Based Conversational Agents

Oct 03, 2026

This study addresses the lack of collaborative visual canvas interaction in voice-based agents by proposing VoCa, a system that enables dynamic coordination between auditory and visual information. Methodologically, we construct a voice-canvas interaction design space and employ user observations, design workshops, and prototype experiments to align spoken dialogue with canvas object creation, annotation, and attention guidance, thereby optimizing multi-turn interaction experiences. The research validates the interactive potential of cross-modal collaboration and reveals core challenges in coordinating “saying” and “showing.” Ultimately, this work provides both theoretical foundations and practical guidelines for interface design and interaction paradigms in multimodal intelligent agents.

0 citationsRead paper

Risk-Averse Welfare Maximization via Marginal Treatment Effects

Sep 24, 2026

This study addresses the challenge of risk-averse treatment assignment when individuals self-select based on unobserved characteristics. It proposes a planner’s optimization framework integrating marginal treatment effects with coherent risk measures, extending endogenous selection models to risk-averse settings. This formulation unifies perspectives on uncertainty aversion, distributional robustness, and worst-case welfare while encompassing the risk-neutral case as a special instance. Leveraging Kusuoka’s representation theorem, the authors characterize optimal assignment rules and establish finite-sample regret bounds for empirical policy learning. Theoretically, this work expands the frontiers of policy learning under causal inference. Empirically, an application to Card (1995) demonstrates that incorporating risk aversion substantially alters optimal college enrollment policies, yielding economically significant implications.

0 citationsRead paper

Founder Backgrounds and Startup Funding: Evidence from Y Combinator

Dec 15, 2025

This study investigates how founders’ backgrounds influence startup fundraising. Using funding data from 4,323 startups admitted to Y Combinator (2005–2024), we estimate an OLS model with batch fixed effects, integrating matched firm-level data from S&P Global and YC’s internal records, and conduct multiple robustness checks. Results show that co-founder count is the most robust predictor: each additional co-founder increases fundraising success by approximately 21%. In contrast, individual credentials—such as prior employment at FAANG firms or elite academic degrees—exhibit negligible explanatory power (collectively accounting for <4% of variance), with unstable coefficients prone to sign reversal. This is the first study to demonstrate, within an elite accelerator cohort, that team size dominates individual pedigree in both statistical robustness and economic significance. Our findings challenge the “star founder” narrative and underscore that team composition—not individual prestige—is the primary determinant of capital acquisition.

0 citationsRead paper

Cooperation as Black Box: Conceptual Fluctuation and Diagnostic Tools for Misalignment in MAS

Jun 28, 2025

In multi-agent systems, conflation of “cooperation” and “coordination,” coupled with moralized misinterpretations, induces semantic ambiguity and normative projection during design—leading to meaning-level misalignment. This paper introduces the “Misalignment Mosaic” diagnostic framework, which systematically identifies latent semantic deviations across four dimensions: terminological inconsistency, concept-to-code decay, moralization of cooperation, and interpretive ambiguity. Innovatively treating “meaning itself” as a source of misalignment, the framework employs the Rabbit-Duck illusion as an analogy to expose the perspective-dependence of behavioral interpretation. As a qualitative tool, it generalizes to other overloaded concepts—including alignment and autonomy—and successfully deconstructs the black-boxing of cooperation. The framework provides an actionable, concept-level diagnostic methodology for multi-agent systems, significantly enhancing semantic consistency and interpretability.

0 citationsRead paper

Recent publications

Latest Papers

VoCa: Designing Speech-Canvas Interaction for Voice-Based Conversational Agents

Oct 03, 2026

This study addresses the lack of collaborative visual canvas interaction in voice-based agents by proposing VoCa, a system that enables dynamic coordination between auditory and visual information. Methodologically, we construct a voice-canvas interaction design space and employ user observations, design workshops, and prototype experiments to align spoken dialogue with canvas object creation, annotation, and attention guidance, thereby optimizing multi-turn interaction experiences. The research validates the interactive potential of cross-modal collaboration and reveals core challenges in coordinating “saying” and “showing.” Ultimately, this work provides both theoretical foundations and practical guidelines for interface design and interaction paradigms in multimodal intelligent agents.

0 citationsRead paper

Risk-Averse Welfare Maximization via Marginal Treatment Effects

Sep 24, 2026

This study addresses the challenge of risk-averse treatment assignment when individuals self-select based on unobserved characteristics. It proposes a planner’s optimization framework integrating marginal treatment effects with coherent risk measures, extending endogenous selection models to risk-averse settings. This formulation unifies perspectives on uncertainty aversion, distributional robustness, and worst-case welfare while encompassing the risk-neutral case as a special instance. Leveraging Kusuoka’s representation theorem, the authors characterize optimal assignment rules and establish finite-sample regret bounds for empirical policy learning. Theoretically, this work expands the frontiers of policy learning under causal inference. Empirically, an application to Card (1995) demonstrates that incorporating risk aversion substantially alters optimal college enrollment policies, yielding economically significant implications.

0 citationsRead paper

Founder Backgrounds and Startup Funding: Evidence from Y Combinator

Dec 15, 2025

This study investigates how founders’ backgrounds influence startup fundraising. Using funding data from 4,323 startups admitted to Y Combinator (2005–2024), we estimate an OLS model with batch fixed effects, integrating matched firm-level data from S&P Global and YC’s internal records, and conduct multiple robustness checks. Results show that co-founder count is the most robust predictor: each additional co-founder increases fundraising success by approximately 21%. In contrast, individual credentials—such as prior employment at FAANG firms or elite academic degrees—exhibit negligible explanatory power (collectively accounting for <4% of variance), with unstable coefficients prone to sign reversal. This is the first study to demonstrate, within an elite accelerator cohort, that team size dominates individual pedigree in both statistical robustness and economic significance. Our findings challenge the “star founder” narrative and underscore that team composition—not individual prestige—is the primary determinant of capital acquisition.

0 citationsRead paper

Cooperation as Black Box: Conceptual Fluctuation and Diagnostic Tools for Misalignment in MAS

Jun 28, 2025

In multi-agent systems, conflation of “cooperation” and “coordination,” coupled with moralized misinterpretations, induces semantic ambiguity and normative projection during design—leading to meaning-level misalignment. This paper introduces the “Misalignment Mosaic” diagnostic framework, which systematically identifies latent semantic deviations across four dimensions: terminological inconsistency, concept-to-code decay, moralization of cooperation, and interpretive ambiguity. Innovatively treating “meaning itself” as a source of misalignment, the framework employs the Rabbit-Duck illusion as an analogy to expose the perspective-dependence of behavioral interpretation. As a qualitative tool, it generalizes to other overloaded concepts—including alignment and autonomy—and successfully deconstructs the black-boxing of cooperation. The framework provides an actionable, concept-level diagnostic methodology for multi-agent systems, significantly enhancing semantic consistency and interpretability.

0 citationsRead paper