Selected work

Representative Papers

A Study of the Plausibility of Attention between RNN Encoders in Natural Language Inference

Dec 01, 2021International Conference on Machine Learning and Applications

Cross-attention maps in natural language inference (NLI) are widely assumed to be interpretable, yet their actual capacity to reveal sentence comparison and logical reasoning processes remains inadequately evaluated—particularly due to the high cost and limited scale of human-annotated explanations. Method: This work introduces, for the first time in NLI, a heuristic rule-based automatic explanation annotation method to overcome these bottlenecks, and conducts a comparative analysis between human and heuristic annotations on the eSNLI dataset. Contribution/Results: Experiments show strong positive correlation (ρ > 0.6) between heuristic and human annotations, validating the former’s utility for explanation quality evaluation. In contrast, raw RNN cross-attention weights exhibit only weak correlation (ρ ≈ 0.2) with human-validated explanations, exposing their limited interpretability. This study establishes a new benchmark, methodology, and empirical insight for assessing attention mechanism interpretability in NLI.

5 citationsRead paper

Write on Paper and Get the Online Digital Trace:\newline A New Era for Handwriting

Sep 11, 2026

本文提出了一种结合数字笔和先进人工智能算法的创新方案，以在普通纸上书写时捕捉手写轨迹，实现无延迟且低成本的手写数字化。

0 citationsRead paper

Quasi-static analysis of passive stability in a novel underactuated multi-finger hand

Sep 10, 2026

本文提出了一种准静态分析方法，用于评估欠驱动多指手的被动稳定性，并通过一种新的集成了差动弹簧滑块机制的三指手架构，研究了不同物体尺寸下的稳定抓取配置。

0 citationsRead paper

REDSI: Addressing the Reproducibility and Evaluation Consistency of Differentiable Search Indexing for Document Retrieval

Sep 08, 2026

本文通过引入ReDSI，一个支持所有三种文档标识符类型的开源实现及规范化的NQ320K构建流程，解决了不同可微搜索索引框架的再现性和评估一致性问题。

0 citationsRead paper

Online Handwriting Trajectory Reconstruction from Kinematic Sensors using Temporal Convolutional Network

Jul 29, 2026

This work addresses the challenge of inaccurate handwriting trajectory reconstruction when writing on arbitrary surfaces by proposing a novel method that leverages motion sensor signals to recover online handwriting trajectories. The approach integrates dynamic time warping (DTW) to align asynchronous multimodal signals and introduces a dedicated temporal convolutional network (TCN) architecture for high-fidelity trajectory reconstruction. Key contributions include the creation of the first publicly available multimodal handwriting trajectory benchmark dataset and an end-to-end reconstruction pipeline that synergistically combines DTW-based preprocessing with TCN modeling. Experimental results demonstrate that the proposed method significantly outperforms existing state-of-the-art techniques in both qualitative and quantitative evaluations, achieving substantial improvements in trajectory recovery accuracy.

0 citationsRead paper

Recent publications

Latest Papers

Write on Paper and Get the Online Digital Trace:\newline A New Era for Handwriting

Sep 11, 2026

本文提出了一种结合数字笔和先进人工智能算法的创新方案，以在普通纸上书写时捕捉手写轨迹，实现无延迟且低成本的手写数字化。

0 citationsRead paper

Quasi-static analysis of passive stability in a novel underactuated multi-finger hand

Sep 10, 2026

本文提出了一种准静态分析方法，用于评估欠驱动多指手的被动稳定性，并通过一种新的集成了差动弹簧滑块机制的三指手架构，研究了不同物体尺寸下的稳定抓取配置。

0 citationsRead paper

REDSI: Addressing the Reproducibility and Evaluation Consistency of Differentiable Search Indexing for Document Retrieval

Sep 08, 2026

本文通过引入ReDSI，一个支持所有三种文档标识符类型的开源实现及规范化的NQ320K构建流程，解决了不同可微搜索索引框架的再现性和评估一致性问题。

0 citationsRead paper

Online Handwriting Trajectory Reconstruction from Kinematic Sensors using Temporal Convolutional Network

Jul 29, 2026

This work addresses the challenge of inaccurate handwriting trajectory reconstruction when writing on arbitrary surfaces by proposing a novel method that leverages motion sensor signals to recover online handwriting trajectories. The approach integrates dynamic time warping (DTW) to align asynchronous multimodal signals and introduces a dedicated temporal convolutional network (TCN) architecture for high-fidelity trajectory reconstruction. Key contributions include the creation of the first publicly available multimodal handwriting trajectory benchmark dataset and an end-to-end reconstruction pipeline that synergistically combines DTW-based preprocessing with TCN modeling. Experimental results demonstrate that the proposed method significantly outperforms existing state-of-the-art techniques in both qualitative and quantitative evaluations, achieving substantial improvements in trajectory recovery accuracy.

0 citationsRead paper

Domain adaptation for handwriting trajectory reconstruction from IMU sensors

Jul 29, 2026

This study addresses the challenge of limited cross-population generalization in IMU-based handwriting trajectory reconstruction, caused by distributional discrepancies in inertial signals between adults and children due to differences in writing speed and confidence. To bridge this gap, the work introduces domain adaptation into the task for the first time, learning a unified intermediate feature representation that effectively aligns signal distributions across user groups. Experimental results demonstrate that the proposed approach significantly outperforms both training from scratch and fine-tuning strategies, achieving more robust and accurate trajectory reconstruction in cross-domain scenarios. This advancement offers a promising new direction toward developing handwriting sensing systems applicable across diverse populations.

0 citationsRead paper