Selected work

Representative Papers

A Deterministic Evidence Layer for Vision-Language Autism Screening from Naturalistic Home Video

Oct 06, 2026

This study addresses the decision instability of vision-language models (VLMs) in autism screening and the scarcity of early-stage expert resources by proposing a grounding-aware, certainty-scoring architecture. By freezing VLM parameters, the method constructs a deterministic evidence layer through timestamped event extraction and textual confidence calibration, effectively decoupling perception from decision-making. Furthermore, a weighted evidence scoring algorithm is designed to achieve stable and interpretable risk stratification, enabling decisions to be decomposed into feature-level contributions with support for rescoring. Evaluated on home videos, the model attains an AUC of 85.1% and an accuracy of 86.0%, with a fully correct annotation rate of 74.4%. Notably, it yields zero false positives among typically developing children while significantly reducing prediction variance.

0 citationsRead paper

Adaptive Co-Serving LLM Watermarking on Modern Inference Engines

Oct 02, 2026

Existing watermarking techniques for large language models are decoupled from inference engines, resulting in high latency, substantial overhead, and deployment challenges. This work proposes SWIFT, a framework that pioneers the co-design of watermark generation and inference infrastructure. By deeply integrating watermark embedding into the vLLM inference backend through instruction-guided candidate generation, asynchronous co-serving, and adaptive scheduling, SWIFT effectively eliminates redundant computation and enhances cache reuse. Experimental results demonstrate that the proposed framework achieves 99.65% detection accuracy and 97.7% robustness against adversarial attacks while preserving a text utility score of 4.87. Furthermore, SWIFT reduces end-to-end latency by 5.9× compared to baseline methods, offering a highly efficient and practical solution for deployable LLM watermarking.

0 citationsRead paper

Validating Memory-Optimal Transformer Kernels on Real Hardware: From Formal Derivation to Measured Performance Across Two HPC Clusters

Sep 27, 2026

This study addresses the significant performance gap between theoretically optimal memory-efficient Transformer kernels and their practical hardware deployment. We propose a method grounded in Mathematics of Arrays (MoA) that models hardware adaptation as Operand Normal Form (ONF) rewriting under a fixed Disjunctive Normal Form (DNF), enabling migration to novel architectures without re-deriving correctness guarantees. Validation is conducted via PyTorch integration and multi-platform profiling across two HPC clusters. The proposed approach achieves up to 2.5× speedup, resolves a regression involving GPU atomic races, quantifies NUMA topology penalties reaching 535×, and identifies the root cause of C/Fortran cross-platform performance inversion.

0 citationsRead paper

ObjectSplat: Improving Mesh Fidelity and Interactivity for 3D Scenes via Object-Level Mesh Splatting

Aug 31, 2026

本文提出了一种通过对象级网格喷溅的方法，解决3D场景重建中缺乏对象级结构及纹理污染问题，提高了网格保真度和新颖视角合成质量。

0 citationsRead paper

Supervised Mixed-Frequency Learning for Macro-Financial Forecasting When Factors are Weak

Aug 12, 2026

This study addresses the limitations of traditional MIDAS models, which rely on strong factor assumptions and underperform in macro-financial forecasting when factors are weak. To overcome this, the authors propose the SsPCA-MIDAS model, which integrates supervised scaled principal component analysis (SsPCA) into the mixed-data sampling framework. This approach achieves, for the first time under weak factor conditions, consistent estimation and asymptotic normality, thereby enabling valid statistical inference. Moreover, the model can be combined with machine learning techniques such as Boosting to enhance predictive accuracy. Empirical results demonstrate that SsPCA-MIDAS significantly outperforms existing methods in forecasting key U.S. macroeconomic and financial indicators—including GDP growth, inflation, unemployment, asset prices, and volatility—and successfully identifies economically meaningful predictors.

0 citationsRead paper

Recent publications

Latest Papers

A Deterministic Evidence Layer for Vision-Language Autism Screening from Naturalistic Home Video

Oct 06, 2026

This study addresses the decision instability of vision-language models (VLMs) in autism screening and the scarcity of early-stage expert resources by proposing a grounding-aware, certainty-scoring architecture. By freezing VLM parameters, the method constructs a deterministic evidence layer through timestamped event extraction and textual confidence calibration, effectively decoupling perception from decision-making. Furthermore, a weighted evidence scoring algorithm is designed to achieve stable and interpretable risk stratification, enabling decisions to be decomposed into feature-level contributions with support for rescoring. Evaluated on home videos, the model attains an AUC of 85.1% and an accuracy of 86.0%, with a fully correct annotation rate of 74.4%. Notably, it yields zero false positives among typically developing children while significantly reducing prediction variance.

0 citationsRead paper

Adaptive Co-Serving LLM Watermarking on Modern Inference Engines

Oct 02, 2026

Existing watermarking techniques for large language models are decoupled from inference engines, resulting in high latency, substantial overhead, and deployment challenges. This work proposes SWIFT, a framework that pioneers the co-design of watermark generation and inference infrastructure. By deeply integrating watermark embedding into the vLLM inference backend through instruction-guided candidate generation, asynchronous co-serving, and adaptive scheduling, SWIFT effectively eliminates redundant computation and enhances cache reuse. Experimental results demonstrate that the proposed framework achieves 99.65% detection accuracy and 97.7% robustness against adversarial attacks while preserving a text utility score of 4.87. Furthermore, SWIFT reduces end-to-end latency by 5.9× compared to baseline methods, offering a highly efficient and practical solution for deployable LLM watermarking.

0 citationsRead paper

Validating Memory-Optimal Transformer Kernels on Real Hardware: From Formal Derivation to Measured Performance Across Two HPC Clusters

Sep 27, 2026

This study addresses the significant performance gap between theoretically optimal memory-efficient Transformer kernels and their practical hardware deployment. We propose a method grounded in Mathematics of Arrays (MoA) that models hardware adaptation as Operand Normal Form (ONF) rewriting under a fixed Disjunctive Normal Form (DNF), enabling migration to novel architectures without re-deriving correctness guarantees. Validation is conducted via PyTorch integration and multi-platform profiling across two HPC clusters. The proposed approach achieves up to 2.5× speedup, resolves a regression involving GPU atomic races, quantifies NUMA topology penalties reaching 535×, and identifies the root cause of C/Fortran cross-platform performance inversion.

0 citationsRead paper

ObjectSplat: Improving Mesh Fidelity and Interactivity for 3D Scenes via Object-Level Mesh Splatting

Aug 31, 2026

本文提出了一种通过对象级网格喷溅的方法，解决3D场景重建中缺乏对象级结构及纹理污染问题，提高了网格保真度和新颖视角合成质量。

0 citationsRead paper

Supervised Mixed-Frequency Learning for Macro-Financial Forecasting When Factors are Weak

Aug 12, 2026

This study addresses the limitations of traditional MIDAS models, which rely on strong factor assumptions and underperform in macro-financial forecasting when factors are weak. To overcome this, the authors propose the SsPCA-MIDAS model, which integrates supervised scaled principal component analysis (SsPCA) into the mixed-data sampling framework. This approach achieves, for the first time under weak factor conditions, consistent estimation and asymptotic normality, thereby enabling valid statistical inference. Moreover, the model can be combined with machine learning techniques such as Boosting to enhance predictive accuracy. Empirical results demonstrate that SsPCA-MIDAS significantly outperforms existing methods in forecasting key U.S. macroeconomic and financial indicators—including GDP growth, inflation, unemployment, asset prices, and volatility—and successfully identifies economically meaningful predictors.

0 citationsRead paper