Selected work

Representative Papers

Vision-Language Models Meet Meteorology: Developing Models for Extreme Weather Events Detection with Heatmaps

Jun 14, 2024arXiv.org

Existing vision-language models (VLMs) exhibit color perception bias and imprecise spatial localization when interpreting meteorological heatmaps, leading to unreliable explanations for extreme weather event detection (EWED). To address this, we formulate EWED as a vision-language question answering (VQA) task and introduce three key contributions: (1) ClimateIQA—the first domain-specific VQA dataset for meteorology; (2) SPOT, a novel algorithm that enhances precise localization of heatmap color boundaries and critical regions; and (3) Climate-Zoo, a family of meteorology-specialized VLMs. Experiments demonstrate that our approach elevates EWED accuracy from 0% to over 90%, substantially outperforming general-purpose VLMs. All datasets, source code, and pretrained models are publicly released, establishing a reproducible benchmark and foundational infrastructure for AI-driven meteorology.

3 citationsRead paper

Hawkes-Driven OTC Market Making: Volterra-Riccati Approximation

Aug 03, 2026

This study addresses the path-dependency challenges in over-the-counter (OTC) market making arising from request-for-quote (RFQ) arrivals driven by Hawkes processes. The authors develop a market-making model based on general Hawkes kernels, employing forward curves of conditional intensities to capture the history-dependent and long-memory dynamics of order flow. They innovatively introduce a multi-level Volterra–Riccati approximation framework that enables efficient state-feedback control while preserving the memory structure inherent in Hawkes processes. This approach is the first to effectively incorporate long-memory RFQ dynamics into optimal market-making strategies: under exponential Hawkes specifications, it closely approximates the exact solution and substantially outperforms memoryless Poisson benchmarks; in power-law long-memory settings, it successfully translates RFQ bursts into sustained quote skewness, significantly enhancing inventory and profit-and-loss risk management.

0 citationsRead paper

Optimal Execution with Passive Market Impact

Jul 30, 2026

This study addresses the trade-off among execution probability, adverse selection, and opportunity cost in limit order trading by proposing a mesoscale optimal passive execution strategy. Embedding two empirically observed microstructural features—namely, the exponential decay of limit order fill probability with quote distance and the short-term linear price response to order flow imbalance—into a stochastic control framework, the work derives for the first time a passively induced market impact rate exhibiting exponential decay and solves for the corresponding optimal liquidation policy. The model is validated on both NASDAQ equity and foreign exchange data and extends naturally to settings involving heterogeneous decay rates, instantaneous impact, and target execution schedules, thereby establishing a theoretical foundation and practical mechanism for tactical passive execution.

0 citationsRead paper

Gaussian Boson Sampling for Asset Clustering in Statistical Arbitrage Portfolios

Jul 21, 2026

This study addresses the challenge of identifying co-moving assets and constructing robust statistical arbitrage portfolios in highly volatile markets by proposing a quantum graph clustering approach based on Gaussian Boson Sampling (GBS). The method maps residual correlations of S&P 500 assets into an adjacency matrix amenable to GBS processing and employs a rolling-window framework to dynamically construct market-neutral portfolios. Key innovations include the introduction of a novel GBS Roots algorithm and the first integration of coherent displacement techniques to mitigate photon loss, substantially enhancing clustering performance under high-loss conditions. Empirical results demonstrate that the proposed approach significantly boosts excess returns during periods of elevated market volatility and maintains consistent outperformance across a wide range of photon loss levels.

0 citationsRead paper

Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents

Jul 14, 2026

This work addresses the challenge that existing self-evolving agent systems rely on reliable evaluation metrics, which are often unavailable in real-world settings due to the absence of trustworthy supervisory signals. To overcome this limitation, the paper proposes a self-improvement framework that operates without pre-existing high-quality scoring standards by co-evolving evaluation metrics and a skill library to achieve robust autonomous enhancement. Key innovations include an evolvable, transparent metric mechanism and a dual-ratchet co-evolutionary architecture, integrated with an evolutionary metric loop driven by ensembles of flaw detectors, anchor reference set training, unsupervised output consensus regularization, and an external auditing mechanism. Evaluated on code generation, enterprise-grade text-to-SQL, and reference-free report generation tasks, the approach attains 88–110% of the performance of ground-truth-metric-driven methods and demonstrates verified safety and efficacy under independent human evaluation.

0 citationsRead paper

Recent publications

Latest Papers

Hawkes-Driven OTC Market Making: Volterra-Riccati Approximation

Aug 03, 2026

This study addresses the path-dependency challenges in over-the-counter (OTC) market making arising from request-for-quote (RFQ) arrivals driven by Hawkes processes. The authors develop a market-making model based on general Hawkes kernels, employing forward curves of conditional intensities to capture the history-dependent and long-memory dynamics of order flow. They innovatively introduce a multi-level Volterra–Riccati approximation framework that enables efficient state-feedback control while preserving the memory structure inherent in Hawkes processes. This approach is the first to effectively incorporate long-memory RFQ dynamics into optimal market-making strategies: under exponential Hawkes specifications, it closely approximates the exact solution and substantially outperforms memoryless Poisson benchmarks; in power-law long-memory settings, it successfully translates RFQ bursts into sustained quote skewness, significantly enhancing inventory and profit-and-loss risk management.

0 citationsRead paper

Optimal Execution with Passive Market Impact

Jul 30, 2026

This study addresses the trade-off among execution probability, adverse selection, and opportunity cost in limit order trading by proposing a mesoscale optimal passive execution strategy. Embedding two empirically observed microstructural features—namely, the exponential decay of limit order fill probability with quote distance and the short-term linear price response to order flow imbalance—into a stochastic control framework, the work derives for the first time a passively induced market impact rate exhibiting exponential decay and solves for the corresponding optimal liquidation policy. The model is validated on both NASDAQ equity and foreign exchange data and extends naturally to settings involving heterogeneous decay rates, instantaneous impact, and target execution schedules, thereby establishing a theoretical foundation and practical mechanism for tactical passive execution.

0 citationsRead paper

Gaussian Boson Sampling for Asset Clustering in Statistical Arbitrage Portfolios

Jul 21, 2026

This study addresses the challenge of identifying co-moving assets and constructing robust statistical arbitrage portfolios in highly volatile markets by proposing a quantum graph clustering approach based on Gaussian Boson Sampling (GBS). The method maps residual correlations of S&P 500 assets into an adjacency matrix amenable to GBS processing and employs a rolling-window framework to dynamically construct market-neutral portfolios. Key innovations include the introduction of a novel GBS Roots algorithm and the first integration of coherent displacement techniques to mitigate photon loss, substantially enhancing clustering performance under high-loss conditions. Empirical results demonstrate that the proposed approach significantly boosts excess returns during periods of elevated market volatility and maintains consistent outperformance across a wide range of photon loss levels.

0 citationsRead paper

Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents

Jul 14, 2026

This work addresses the challenge that existing self-evolving agent systems rely on reliable evaluation metrics, which are often unavailable in real-world settings due to the absence of trustworthy supervisory signals. To overcome this limitation, the paper proposes a self-improvement framework that operates without pre-existing high-quality scoring standards by co-evolving evaluation metrics and a skill library to achieve robust autonomous enhancement. Key innovations include an evolvable, transparent metric mechanism and a dual-ratchet co-evolutionary architecture, integrated with an evolutionary metric loop driven by ensembles of flaw detectors, anchor reference set training, unsupervised output consensus regularization, and an external auditing mechanism. Evaluated on code generation, enterprise-grade text-to-SQL, and reference-free report generation tasks, the approach attains 88–110% of the performance of ground-truth-metric-driven methods and demonstrates verified safety and efficacy under independent human evaluation.

0 citationsRead paper

Strategic OTC market making with reputation feedback

Jul 13, 2026

This study investigates how market makers in electronic over-the-counter trading balance immediate spread-based profits against the long-term value of client flow. To this end, the authors develop a stochastic control model that explicitly incorporates reputation dynamics through a performance-based order flow feedback mechanism—where future order arrival intensity depends on both request-for-quote (RFQ) win rates and streaming execution rates. By integrating stochastic dynamic programming with optimal control theory, the model naturally yields an optimal strategy characterized by alternating phases of reputation accumulation and profit realization. The analysis further reveals that, even under a single market maker, multiple stable client-flow equilibria can emerge, thereby elucidating the dynamic trade-off between building reputation and monetizing franchise value.

0 citationsRead paper