Selected work

Representative Papers

Can Open-Weight Models Compete on Financial Text Comprehension?

Aug 09, 2026

This study addresses the lack of systematic evaluation of open-source large language models on real-world financial text understanding tasks by extending the Financial Touchstone benchmark, which now comprises 2,967 question-answer triplets derived from 495 international annual reports. The authors conduct a comprehensive assessment of 20 prominent models, including GLM-4.7, GLM-5, Kimi K2.6, DeepSeek V3.2, and leading closed-source counterparts. Notably, they find that certain non-reasoning open-source models—such as Kimi K2.6 and GLM-5—can match or even surpass some closed-source models in financial comprehension. The study also reveals that Chinese-language models often unjustifiably refuse legitimate queries due to content filtering mechanisms, with refusal behavior influenced by access pathways. Among evaluated models, Claude Opus 4.6 achieves the highest accuracy (88.4%), while Gemini 2.5 Pro exhibits the lowest hallucination rate (0.08%). Information retrieval errors account for 48.9% of failures. The full dataset and evaluation framework are publicly released.

0 citationsRead paper

Robust Control under Stationary Ambiguity

Aug 05, 2026

This work addresses the degradation of long-term robustness in learned control policies when deployed on real systems, a problem arising from existing simulators’ failure to properly model uncertainty in parameter estimation. To mitigate this, the authors propose a “stationary ambiguity” modeling principle that constructs a simulator with state-dependent parameter uncertainty that does not systematically deteriorate over time. By integrating stochastic parameter sampling, state-dependent ambiguity sets, Bayesian filtering, and reinforcement learning, the approach ensures a stationary latent state filtering process. Evaluated on a financial hedging task, policies trained with this method demonstrate sustained robustness and superior performance on real market data, validating the efficacy of the proposed framework.

0 citationsRead paper

TimeRLM: Recursive Language Models Enable Precise Anomaly Localization in Long-Context Time-Series

Aug 04, 2026

This work addresses the challenging task of anomaly localization in long-context time series, where critical evidence is often embedded within long-range dependencies of high-frequency data. To tackle this, we propose TimeRLM—the first recurrent language model specifically designed for time series anomaly localization—integrating code-driven signal operations, vision-assisted comprehension, an external context querying mechanism, and reinforcement learning-based post-training. We also introduce AnomalyXL, a synthetic benchmark for systematic evaluation. Experimental results demonstrate that TimeRLM achieves an IoU of 0.682 and a classification-with-evidence accuracy of 0.745 on AnomalyXL-Localize, substantially outperforming existing time series language models, while maintaining or even improving performance on real-world ECG, sleep, and software monitoring datasets.

0 citationsRead paper

Universality and Approximation Rates of Graph Neural Networks with Random Features

Jul 29, 2026

This work investigates the expressive power of graph neural networks, specifically their ability to approximate arbitrary permutation-invariant or equivariant functions. To this end, the authors propose Permutation-Equivariant Neural Networks with partial random node features (PENN) and establish, for the first time, a universal approximation theorem for such architectures: on fixed-size directed graphs, PENN can probabilistically approximate any measurable permutation-invariant or equivariant function to arbitrary precision. Moreover, for functions that are $k$-times continuously differentiable with $k \geq 2$, the study derives a quantitative upper bound on the approximation error in terms of network depth and the number of non-zero weights, thereby revealing an explicit relationship between model complexity and approximation rate.

0 citationsRead paper

A Survey on Code Equivalence: The State-of-the-Art and Open Questions

Jul 14, 2026

This study systematically surveys the current state of research on code equivalence problems and their principal variants, integrating diverse problem formulations, algorithmic techniques, and hardness assumptions into a unified theoretical framework. Through comprehensive literature review and categorization, combined with insights from cryptographic complexity theory and algebraic coding theory, the work rigorously evaluates the efficacy and parameter regimes of existing algorithms and attack strategies. The analysis not only clarifies the present boundaries of feasible techniques and inherent methodological limitations but also explicitly identifies several critical open problems for the first time, thereby offering a clear roadmap for future theoretical advances and practical applications in code-based cryptography.

0 citationsRead paper

Recent publications

Latest Papers

Can Open-Weight Models Compete on Financial Text Comprehension?

Aug 09, 2026

This study addresses the lack of systematic evaluation of open-source large language models on real-world financial text understanding tasks by extending the Financial Touchstone benchmark, which now comprises 2,967 question-answer triplets derived from 495 international annual reports. The authors conduct a comprehensive assessment of 20 prominent models, including GLM-4.7, GLM-5, Kimi K2.6, DeepSeek V3.2, and leading closed-source counterparts. Notably, they find that certain non-reasoning open-source models—such as Kimi K2.6 and GLM-5—can match or even surpass some closed-source models in financial comprehension. The study also reveals that Chinese-language models often unjustifiably refuse legitimate queries due to content filtering mechanisms, with refusal behavior influenced by access pathways. Among evaluated models, Claude Opus 4.6 achieves the highest accuracy (88.4%), while Gemini 2.5 Pro exhibits the lowest hallucination rate (0.08%). Information retrieval errors account for 48.9% of failures. The full dataset and evaluation framework are publicly released.

0 citationsRead paper

Robust Control under Stationary Ambiguity

Aug 05, 2026

This work addresses the degradation of long-term robustness in learned control policies when deployed on real systems, a problem arising from existing simulators’ failure to properly model uncertainty in parameter estimation. To mitigate this, the authors propose a “stationary ambiguity” modeling principle that constructs a simulator with state-dependent parameter uncertainty that does not systematically deteriorate over time. By integrating stochastic parameter sampling, state-dependent ambiguity sets, Bayesian filtering, and reinforcement learning, the approach ensures a stationary latent state filtering process. Evaluated on a financial hedging task, policies trained with this method demonstrate sustained robustness and superior performance on real market data, validating the efficacy of the proposed framework.

0 citationsRead paper

TimeRLM: Recursive Language Models Enable Precise Anomaly Localization in Long-Context Time-Series

Aug 04, 2026

This work addresses the challenging task of anomaly localization in long-context time series, where critical evidence is often embedded within long-range dependencies of high-frequency data. To tackle this, we propose TimeRLM—the first recurrent language model specifically designed for time series anomaly localization—integrating code-driven signal operations, vision-assisted comprehension, an external context querying mechanism, and reinforcement learning-based post-training. We also introduce AnomalyXL, a synthetic benchmark for systematic evaluation. Experimental results demonstrate that TimeRLM achieves an IoU of 0.682 and a classification-with-evidence accuracy of 0.745 on AnomalyXL-Localize, substantially outperforming existing time series language models, while maintaining or even improving performance on real-world ECG, sleep, and software monitoring datasets.

0 citationsRead paper

Universality and Approximation Rates of Graph Neural Networks with Random Features

Jul 29, 2026

This work investigates the expressive power of graph neural networks, specifically their ability to approximate arbitrary permutation-invariant or equivariant functions. To this end, the authors propose Permutation-Equivariant Neural Networks with partial random node features (PENN) and establish, for the first time, a universal approximation theorem for such architectures: on fixed-size directed graphs, PENN can probabilistically approximate any measurable permutation-invariant or equivariant function to arbitrary precision. Moreover, for functions that are $k$-times continuously differentiable with $k \geq 2$, the study derives a quantitative upper bound on the approximation error in terms of network depth and the number of non-zero weights, thereby revealing an explicit relationship between model complexity and approximation rate.

0 citationsRead paper

A Survey on Code Equivalence: The State-of-the-Art and Open Questions

Jul 14, 2026

This study systematically surveys the current state of research on code equivalence problems and their principal variants, integrating diverse problem formulations, algorithmic techniques, and hardness assumptions into a unified theoretical framework. Through comprehensive literature review and categorization, combined with insights from cryptographic complexity theory and algebraic coding theory, the work rigorously evaluates the efficacy and parameter regimes of existing algorithms and attack strategies. The analysis not only clarifies the present boundaries of feasible techniques and inherent methodological limitations but also explicitly identifies several critical open problems for the first time, thereby offering a clear roadmap for future theoretical advances and practical applications in code-based cryptography.

0 citationsRead paper