Selected work

Representative Papers

PRISM: Distribution-free Adaptive Computation of Matrix Functions for Accelerating Neural Network Training

Jan 29, 2026

This work addresses the computational inefficiency of matrix functions—such as square roots, inverse roots, and orthogonalization—in neural network training, which stems from traditional iterative methods’ reliance on prior spectral information and their inability to adapt to dynamically changing matrix spectra. The authors propose PRISM, a novel framework that enables adaptive computation of matrix functions without requiring any prior knowledge of the spectrum. PRISM constructs, at each iteration, a polynomial surrogate of the current spectrum using random sketching and relies predominantly on GPU-friendly matrix multiplications. This approach automatically adapts to spectral shifts during training, substantially reducing computational overhead. When integrated into Shampoo and Muon optimizers, PRISM maintains optimization accuracy while significantly decreasing both iteration counts and wall-clock runtime.

1 citationsRead paper

Accelerating scientific discovery with the common task framework

Nov 06, 2025

The scientific and engineering communities lack unified, reproducible benchmarks for evaluating AI/ML methods in dynamical systems modeling. Method: This paper introduces the Common Task Framework (CTF), a general-purpose framework targeting multiple scientific objectives—including prediction, state reconstruction, generalization, and control—under realistic constraints of limited data and noisy measurements. CTF establishes standardized datasets, objective evaluation metrics, and an open benchmarking platform. Contribution/Results: CTF enables the first cross-disciplinary, physics-constrained comparison of system identification and machine learning algorithms, facilitating rapid iterative development and integration. Experimental results demonstrate that CTF significantly improves model development efficiency and deployment reliability, thereby addressing a critical gap in AI evaluation frameworks oriented toward scientific discovery.

1 citationsRead paper

Toward Integrated Solutions: A Systematic Interdisciplinary Review of Cybergrooming Research

Feb 18, 2025arXiv.org

Cybergrooming research suffers from a persistent disciplinary divide: social science studies emphasize behavioral insights, while computational approaches focus on detection techniques—limiting interdisciplinary integration and real-world intervention efficacy. This study conducts the first systematic, cross-disciplinary literature review following the PRISMA framework, bridging qualitative behavioral analysis and machine learning–based detection models. Through integrated qualitative coding, model evaluation, and multidimensional metric comparison, we identify three critical bottlenecks: poor data quality, inconsistent evaluation standards, and lack of cultural adaptation. We propose a unified research paradigm that incorporates cultural sensitivity, bias mitigation, and balanced evaluation criteria. The resulting methodology is reproducible, extensible, and empirically grounded—providing both theoretical foundations and actionable pathways for developing collaborative, evidence-based online protection systems for minors. (149 words)

1 citationsRead paper

LLM Inference in a Flash!

Sep 14, 2026

为解决大语言模型推理面临的内存带宽限制及硬件挑战，本文提出了一种基于整数量化的计算方法，并设计了字典式KV缓存压缩策略，以减少数据传输并提高Flash计算效率。

0 citationsRead paper

Reconceptualizing Age Assurance as a Sociotechnical Problem: Connecting Evidence, Evaluation, Claims, and Decisions

Sep 11, 2026

论文将年龄验证视为社会技术问题，提出年龄保证过程框架，连接证据、评估、声明和决策，而不仅仅是技术准确性。

0 citationsRead paper

Recent publications

Latest Papers

LLM Inference in a Flash!

Sep 14, 2026

为解决大语言模型推理面临的内存带宽限制及硬件挑战，本文提出了一种基于整数量化的计算方法，并设计了字典式KV缓存压缩策略，以减少数据传输并提高Flash计算效率。

0 citationsRead paper

Reconceptualizing Age Assurance as a Sociotechnical Problem: Connecting Evidence, Evaluation, Claims, and Decisions

Sep 11, 2026

论文将年龄验证视为社会技术问题，提出年龄保证过程框架，连接证据、评估、声明和决策，而不仅仅是技术准确性。

0 citationsRead paper

DeepSAGE: Stage-Aware Reinforcement Learning for Structured CBT Counseling Dialogue

Aug 23, 2026

为解决LLM咨询代理缺乏结构化目标导向的问题，提出DeepSAGE框架，结合深度强化学习和阶段意识进行CBT对话。

0 citationsRead paper

Eigenanalysis framework for autoregressive neural emulators of multi-scale chaotic dynamics

Aug 17, 2026

This study addresses the long-term instability and obscure error mechanisms in neural autoregressive modeling of chaotic systems by constructing a feature analysis framework to uncover the dynamical origins of error growth. Through Jacobian spectral analysis and high-order numerical integrators, we establish an a priori stability diagnostic theory that reveals a universal linear error scaling law for integration-constrained models and proposes a stability regularization loss function. Extensive validation across 29 architectures demonstrates that this approach significantly enhances prediction accuracy and dynamical robustness. Consequently, this work provides both theoretical underpinnings and an effective optimization paradigm for neural network-based modeling of chaotic dynamics, bridging the gap between numerical stability theory and deep learning applications in complex system simulation.

0 citationsRead paper

From Chasing Ghosts to Missed Attacks: Perspectives and Perceptions of SOC Practitioners on LLM Integration, Risks, and Readiness

Aug 01, 2026

This study addresses the lack of empirical, practitioner-based insights into the applicability and risks of large language models (LLMs) within real-world security operations center (SOC) workflows. Through semi-structured interviews and interactive scenario simulations with 25 SOC practitioners experienced in LLM use, the research systematically identifies six functional categories and 15 concrete use cases for LLMs in cybersecurity operations from a human-centered perspective, while proposing integration design requirements grounded in operational safety. Findings indicate that LLMs are well-suited for low-level, repetitive tasks such as report automation but face limitations in high-impact activities like incident analysis due to insufficient technical depth and contextual awareness. Although practitioners express caution about overreliance, they broadly support the judicious adoption of LLMs within SOC environments.

0 citationsRead paper