Selected work

Representative Papers

Multi-Person Pose Estimation Evaluation Using Optimal Transportation and Improved Pose Matching

Jul 26, 20252025 19th International Conference on Machine Vision and Applications (MVA)

Existing evaluation metrics for multi-person pose estimation overly rely on the ranking of high-confidence detections while neglecting low-confidence false positives, leading to biased assessments. To address this limitation, this work proposes OCpose, which introduces optimal transport theory into pose evaluation for the first time. By employing a confidence-weighted matching strategy, OCpose achieves globally optimal alignment between detected poses and ground-truth annotations. This approach abandons the conventional reliance on confidence-based ranking and instead fairly balances true positives against false positives, yielding a more comprehensive and unbiased evaluation of model performance. As a result, OCpose significantly enhances the robustness and reasonableness of pose estimation assessment.

1 citationsRead paper

Shaping Human-AI Interactions to Provide Improvement Pathways and Balance Competing Objectives

Aug 06, 2026

This work addresses the challenge that users often adopt gaming strategies against deployed AI systems due to misconceptions about their mechanisms, undermining both personal development and system objectives such as predictive accuracy. To counter this, the study proposes a novel human-AI interaction design principle that systematically integrates the perspectives of both the evaluated individuals and the AI system itself. This approach simultaneously incentivizes genuine self-improvement and preserves system performance. Through theoretical analysis, data-driven modeling, human-subject experiments, and validation on real and semi-synthetic datasets, the research demonstrates that the proposed method effectively calibrates user beliefs, substantially reduces gaming behaviors, fosters meaningful self-enhancement, and maintains the stability of core AI performance metrics. The findings offer a new paradigm for the trustworthy deployment of AI in practice.

0 citationsRead paper

On the Power of Deception in Repeated Games

Jul 25, 2026

This study investigates how a strategic agent can exploit an opponent’s history-dependent prediction mechanism through deception to enhance its own payoff in repeated games. Focusing on count-based learners that employ empirical risk minimization (ERM), the work formalizes deceptive behavior for the first time and introduces a novel metric termed “deception reward.” It reveals structural properties of deception in general-sum games and proves that approximating the optimal deception reward is NP-hard. The proposed approach combines dynamic programming with approximation algorithms, enabling exact optimization when the action space is small or the opponent has short memory, and yielding effective approximations in settings with long horizons or unknown learning rules. Empirical results demonstrate the existence of substantial deception rewards in stochastic games.

0 citationsRead paper

Algorithmic Approaches to Sequential Decision-Making and Social Epistemology

Jul 22, 2026

This study addresses sequential decision-making in humans concerning persistence versus abandonment, and investigates phenomena such as pessimism traps and insufficient ambition arising from behavioral biases and social influence in social cognition. By constructing an enhanced multi-armed bandit model, the work introduces data-driven algorithmic design into sequential decision-making for the first time and provides a formal characterization of pessimism traps. Theoretical analysis yields nearly tight upper and lower bounds on sample complexity under general conditions, demonstrating that polynomially many samples suffice to learn near-optimal policies. Furthermore, the paper proposes a sustainable community intervention mechanism that effectively disrupts pessimism traps, thereby bridging abstract theories in social epistemology with the complexities of real-world decision-making.

0 citationsRead paper

On Incentivized Exploration beyond Bayesianism and Full-Information

Jul 14, 2026

This work addresses the limitations of traditional Bayesian incentive compatibility, which fails when agents possess private information, lack a common prior, or operate in incomplete information environments. The paper proposes a more general incentive exploration framework that dispenses with Bayesian assumptions and requirements of complete information, allowing agents to act according to any undominated strategy. By introducing a novel definition of incentive compatibility that does not rely on a common prior and integrating multi-prior robust optimization with non-Bayesian decision theory, the framework effectively handles action ties. This approach substantially extends the applicability and robustness of incentive-compatible mechanisms under information asymmetry and uncertainty.

0 citationsRead paper

Recent publications

Latest Papers

Shaping Human-AI Interactions to Provide Improvement Pathways and Balance Competing Objectives

Aug 06, 2026

This work addresses the challenge that users often adopt gaming strategies against deployed AI systems due to misconceptions about their mechanisms, undermining both personal development and system objectives such as predictive accuracy. To counter this, the study proposes a novel human-AI interaction design principle that systematically integrates the perspectives of both the evaluated individuals and the AI system itself. This approach simultaneously incentivizes genuine self-improvement and preserves system performance. Through theoretical analysis, data-driven modeling, human-subject experiments, and validation on real and semi-synthetic datasets, the research demonstrates that the proposed method effectively calibrates user beliefs, substantially reduces gaming behaviors, fosters meaningful self-enhancement, and maintains the stability of core AI performance metrics. The findings offer a new paradigm for the trustworthy deployment of AI in practice.

0 citationsRead paper

On the Power of Deception in Repeated Games

Jul 25, 2026

This study investigates how a strategic agent can exploit an opponent’s history-dependent prediction mechanism through deception to enhance its own payoff in repeated games. Focusing on count-based learners that employ empirical risk minimization (ERM), the work formalizes deceptive behavior for the first time and introduces a novel metric termed “deception reward.” It reveals structural properties of deception in general-sum games and proves that approximating the optimal deception reward is NP-hard. The proposed approach combines dynamic programming with approximation algorithms, enabling exact optimization when the action space is small or the opponent has short memory, and yielding effective approximations in settings with long horizons or unknown learning rules. Empirical results demonstrate the existence of substantial deception rewards in stochastic games.

0 citationsRead paper

Algorithmic Approaches to Sequential Decision-Making and Social Epistemology

Jul 22, 2026

This study addresses sequential decision-making in humans concerning persistence versus abandonment, and investigates phenomena such as pessimism traps and insufficient ambition arising from behavioral biases and social influence in social cognition. By constructing an enhanced multi-armed bandit model, the work introduces data-driven algorithmic design into sequential decision-making for the first time and provides a formal characterization of pessimism traps. Theoretical analysis yields nearly tight upper and lower bounds on sample complexity under general conditions, demonstrating that polynomially many samples suffice to learn near-optimal policies. Furthermore, the paper proposes a sustainable community intervention mechanism that effectively disrupts pessimism traps, thereby bridging abstract theories in social epistemology with the complexities of real-world decision-making.

0 citationsRead paper

On Incentivized Exploration beyond Bayesianism and Full-Information

Jul 14, 2026

This work addresses the limitations of traditional Bayesian incentive compatibility, which fails when agents possess private information, lack a common prior, or operate in incomplete information environments. The paper proposes a more general incentive exploration framework that dispenses with Bayesian assumptions and requirements of complete information, allowing agents to act according to any undominated strategy. By introducing a novel definition of incentive compatibility that does not rely on a common prior and integrating multi-prior robust optimization with non-Bayesian decision theory, the framework effectively handles action ties. This approach substantially extends the applicability and robustness of incentive-compatible mechanisms under information asymmetry and uncertainty.

0 citationsRead paper

The Optimal Sample Complexity of Learning Autoregressive Chain-of-Thought

Jul 08, 2026

This work addresses the high sample complexity of autoregressive chain-of-thought learning, where a single error can invalidate an entire trajectory. Within the realizable PAC framework, the paper introduces the “parity dimension”—a rollout-stable refinement of the Daniely–Shalev-Shwartz (DS) dimension—that effectively mitigates the dimensional blow-up inherent in traditional DS dimension when applied to autoregressive unrolling. By integrating PAC theory, parity pseudo-cubes, and combinatorial analysis, the authors establish a sample complexity upper bound of \(O((\text{DSdim}(\mathcal{H}) + \log(1/\delta))/\varepsilon)\). This bound depends only on the complexity of local next-token prediction and is independent of the reasoning horizon; moreover, it is shown to be unimprovable in the worst case.

0 citationsRead paper