Selected work

Representative Papers

Cost-Efficient Theorem Proving via Agent Orchestration in Program Verification

Oct 07, 2026

This study addresses the challenges of large-scale proof obligations and the difficulty of balancing success rates against computational overhead in program verification by proposing the CoCo-Prover framework. This method formulates formal proving as cost-aware meta-level decision-making, employing a bi-level AND/OR hypergraph proof structure and lemma dependency graphs to enable symbolic topological selection. Furthermore, it treats expensive expert invocations as priced services for dynamic agent orchestration and routing. Evaluated in the Lean 4 environment across five benchmarks, CoCo-Prover achieves state-of-the-art solve rates—reaching up to 100%—while reducing computational costs by 30.9% compared to the strongest baseline. Ultimately, this work effectively unifies efficiency and economy in automated theorem proving.

0 citationsRead paper

Empath: Tracing Multi-Level Emotion Dynamics in Crisis Counseling Dialogues

Sep 24, 2026

This study addresses the limitation of static emotion labeling in crisis intervention dialogues, which fails to capture dynamic emotional evolution. To this end, we propose EMPATH, a framework integrating textual affective computing with statistical models to achieve turn-level, transition-probability, and global-prototype emotion dynamics modeling across three granularities—the first of its kind. Our analysis reveals the complex trajectories through which sustained negative emotions transition toward hope during crisis support, while identifying that African American help-seekers exhibit persistent negative affect and heterogeneous recovery pathways. This work overcomes the constraints of traditional static analyses and demonstrates the critical value of multi-granularity dynamic modeling for understanding psychological mechanisms such as grief expression.

0 citationsRead paper

Learning GR(1) Specifications from Traces

Aug 06, 2026

This work addresses the problem of automatically mining synthesizable GR(1) specifications from system execution traces by introducing GR1MINE, an efficient and specialized mining framework. GR1MINE integrates GR(1) temporal skeleton analysis, incremental formula enumeration, and conflict-driven clause learning, leveraging SAT solving to prune redundant search paths and substantially improve both efficiency and coverage of mined specifications. As the first dedicated framework for the GR(1) fragment, GR1MINE achieves a speedup of over 30× compared to general-purpose LTL mining tools on the Syntech benchmark suite and recovers more than twice as many realizable specifications on non-GR(1) instances from SYNTCOMP.

0 citationsRead paper

Recent publications

Latest Papers

Cost-Efficient Theorem Proving via Agent Orchestration in Program Verification

Oct 07, 2026

This study addresses the challenges of large-scale proof obligations and the difficulty of balancing success rates against computational overhead in program verification by proposing the CoCo-Prover framework. This method formulates formal proving as cost-aware meta-level decision-making, employing a bi-level AND/OR hypergraph proof structure and lemma dependency graphs to enable symbolic topological selection. Furthermore, it treats expensive expert invocations as priced services for dynamic agent orchestration and routing. Evaluated in the Lean 4 environment across five benchmarks, CoCo-Prover achieves state-of-the-art solve rates—reaching up to 100%—while reducing computational costs by 30.9% compared to the strongest baseline. Ultimately, this work effectively unifies efficiency and economy in automated theorem proving.

0 citationsRead paper

Empath: Tracing Multi-Level Emotion Dynamics in Crisis Counseling Dialogues

Sep 24, 2026

This study addresses the limitation of static emotion labeling in crisis intervention dialogues, which fails to capture dynamic emotional evolution. To this end, we propose EMPATH, a framework integrating textual affective computing with statistical models to achieve turn-level, transition-probability, and global-prototype emotion dynamics modeling across three granularities—the first of its kind. Our analysis reveals the complex trajectories through which sustained negative emotions transition toward hope during crisis support, while identifying that African American help-seekers exhibit persistent negative affect and heterogeneous recovery pathways. This work overcomes the constraints of traditional static analyses and demonstrates the critical value of multi-granularity dynamic modeling for understanding psychological mechanisms such as grief expression.

0 citationsRead paper

Learning GR(1) Specifications from Traces

Aug 06, 2026

This work addresses the problem of automatically mining synthesizable GR(1) specifications from system execution traces by introducing GR1MINE, an efficient and specialized mining framework. GR1MINE integrates GR(1) temporal skeleton analysis, incremental formula enumeration, and conflict-driven clause learning, leveraging SAT solving to prune redundant search paths and substantially improve both efficiency and coverage of mined specifications. As the first dedicated framework for the GR(1) fragment, GR1MINE achieves a speedup of over 30× compared to general-purpose LTL mining tools on the Syntech benchmark suite and recovers more than twice as many realizable specifications on non-GR(1) instances from SYNTCOMP.

0 citationsRead paper