Selected work

Representative Papers

What Makes a Good TODO Comment?

May 13, 2024ACM Transactions on Software Engineering and Methodology

TODO comments in open-source projects suffer from pervasive low quality (46.7% are vague, information-deficient, or lack practical utility) and chronic lack of resolution, demanding systematic governance. This study first proposes a multidimensional high-quality TODO criterion, empirically derived from lifecycle analysis and management practice comparison of 2,863 TODOs across GitHub’s Top 100 Java repositories. We then develop the first CodeBERT-based fine-tuned model for TODO quality assessment, achieving an F1-score of 0.89 on binary classification. Finally, we deliver actionable writing guidelines and governance recommendations. Our core contributions are threefold: (1) theoretically, the first comprehensive TODO quality assessment framework; (2) methodologically, the first deep learning–driven automated quality identification system; and (3) practically, community-adoptable, evidence-based pathways for improving TODO quality in open-source development.

3 citationsRead paper

B-Spline Embedded Structure Learning for 3D Tooth Segmentation

Aug 17, 2026

为解决3D牙齿分割难题，提出B样条嵌入结构学习方法，通过参数化牙齿拓扑和结构感知动态分类器提高模型准确性与鲁棒性。

0 citationsRead paper

LipCache: A Local Inference Proxy with Certified Caching for Edge Image Classification Service

Aug 13, 2026

This work addresses the challenge of balancing inference efficiency and classification reliability in edge-based image classification under stringent latency and throughput constraints. Existing semantic caching approaches rely on empirical thresholds, which often lead to silent errors near decision boundaries. To overcome this limitation, the authors propose LipCache—a framework that introduces a lightweight GuardNet satisfying Lipschitz continuity, operating alongside an unmodified main model (MainNet). GuardNet maps inputs into a low-dimensional feature space and computes, for each sample, a provably sound reuse radius based on local classification boundaries and the spectral norm of the classifier head. Cache results are reused only when queries fall within this certified ball; otherwise, inference falls back to MainNet. This approach establishes the first per-sample provably correct caching mechanism for image classification, replacing heuristic hit decisions with geometrically certified guarantees that ensure 100% cache consistency. Experiments on CIFAR, Tiny-ImageNet, and SVHN demonstrate up to 1.65× speedup with negligible accuracy loss, and an enhanced GuardNet variant significantly improves hit rates in multi-class scenarios.

0 citationsRead paper

Causality Sum Rules in Conventional Scattering Matrices

Aug 10, 2026

This work addresses the long-standing challenge that conventional scattering matrices struggle to directly encode causality constraints, often relying on auxiliary variables that disconnect theory from experimental measurements. By introducing a domain-delay matrix defined via the earliest arrival time of each channel, the authors restore the causal time origin while preserving real-frequency passivity. Building upon assumptions of analyticity, transparency, and regularity, they construct a Schur-function framework that, for the first time, establishes direct causality sum rules within standard scattering matrices. This bridges fundamental causal theory with measurable scattering data, yielding new constraints on coherent superposition and multi-channel loss. The approach synergistically combines domain-delay transforms, Schur theory, Cayley–Herglotz representations, and singular value analysis, not only recovering Rozanov’s absorber limit and spherical multipole sum rules but also generalizing them to universal causal bounds governing insertion loss, singular-value suppression, and delay–bandwidth trade-offs in conditionally lossless systems.

0 citationsRead paper

ArtAnno: Annotating Implicit Semantics in Artworks through LLM Agent-Driven Bidirectional Human-AI Augmentation

Aug 05, 2026

This study addresses the inefficiency and heavy reliance on manual calibration in semantic annotation of artworks, as well as the lack of effective human–AI collaboration mechanisms in existing tools. To overcome these limitations, the authors propose a bidirectional Human–AI Augmentation framework (BiHAA) that implements a closed-loop system, ArtAnno, leveraging a multi-agent architecture to enable real-time interaction and mutual capability enhancement between human experts and AI during annotation. The framework integrates large language models, proactive intelligent support, interaction-driven trajectory distillation, and experience reuse techniques, introducing a novel dynamic mutual reinforcement mechanism. This approach significantly improves annotation efficiency, reduces the verification burden on non-experts, and facilitates continuous accumulation and reuse of domain-specific knowledge.

0 citationsRead paper

Recent publications

Latest Papers

B-Spline Embedded Structure Learning for 3D Tooth Segmentation

Aug 17, 2026

为解决3D牙齿分割难题，提出B样条嵌入结构学习方法，通过参数化牙齿拓扑和结构感知动态分类器提高模型准确性与鲁棒性。

0 citationsRead paper

LipCache: A Local Inference Proxy with Certified Caching for Edge Image Classification Service

Aug 13, 2026

This work addresses the challenge of balancing inference efficiency and classification reliability in edge-based image classification under stringent latency and throughput constraints. Existing semantic caching approaches rely on empirical thresholds, which often lead to silent errors near decision boundaries. To overcome this limitation, the authors propose LipCache—a framework that introduces a lightweight GuardNet satisfying Lipschitz continuity, operating alongside an unmodified main model (MainNet). GuardNet maps inputs into a low-dimensional feature space and computes, for each sample, a provably sound reuse radius based on local classification boundaries and the spectral norm of the classifier head. Cache results are reused only when queries fall within this certified ball; otherwise, inference falls back to MainNet. This approach establishes the first per-sample provably correct caching mechanism for image classification, replacing heuristic hit decisions with geometrically certified guarantees that ensure 100% cache consistency. Experiments on CIFAR, Tiny-ImageNet, and SVHN demonstrate up to 1.65× speedup with negligible accuracy loss, and an enhanced GuardNet variant significantly improves hit rates in multi-class scenarios.

0 citationsRead paper

Causality Sum Rules in Conventional Scattering Matrices

Aug 10, 2026

This work addresses the long-standing challenge that conventional scattering matrices struggle to directly encode causality constraints, often relying on auxiliary variables that disconnect theory from experimental measurements. By introducing a domain-delay matrix defined via the earliest arrival time of each channel, the authors restore the causal time origin while preserving real-frequency passivity. Building upon assumptions of analyticity, transparency, and regularity, they construct a Schur-function framework that, for the first time, establishes direct causality sum rules within standard scattering matrices. This bridges fundamental causal theory with measurable scattering data, yielding new constraints on coherent superposition and multi-channel loss. The approach synergistically combines domain-delay transforms, Schur theory, Cayley–Herglotz representations, and singular value analysis, not only recovering Rozanov’s absorber limit and spherical multipole sum rules but also generalizing them to universal causal bounds governing insertion loss, singular-value suppression, and delay–bandwidth trade-offs in conditionally lossless systems.

0 citationsRead paper

ArtAnno: Annotating Implicit Semantics in Artworks through LLM Agent-Driven Bidirectional Human-AI Augmentation

Aug 05, 2026

This study addresses the inefficiency and heavy reliance on manual calibration in semantic annotation of artworks, as well as the lack of effective human–AI collaboration mechanisms in existing tools. To overcome these limitations, the authors propose a bidirectional Human–AI Augmentation framework (BiHAA) that implements a closed-loop system, ArtAnno, leveraging a multi-agent architecture to enable real-time interaction and mutual capability enhancement between human experts and AI during annotation. The framework integrates large language models, proactive intelligent support, interaction-driven trajectory distillation, and experience reuse techniques, introducing a novel dynamic mutual reinforcement mechanism. This approach significantly improves annotation efficiency, reduces the verification burden on non-experts, and facilitates continuous accumulation and reuse of domain-specific knowledge.

0 citationsRead paper

Effective and Efficient Context Retrieval via Partial Dependency Graph for Repository-Level Code Generation

Aug 03, 2026

Existing retrieval-augmented generation approaches struggle to effectively capture the dependency context of target functions in repository-scale code generation: similarity-based retrieval ignores dependencies, while static dependency graphs are costly to construct and lack flexibility. Inspired by human developer behavior, this work proposes DyRetriever—a dynamic retrieval mechanism that leverages the semantic reasoning capabilities of large language models to perform multi-hop traversal along dependency paths starting from entry functions, constructing and discarding local dependency graphs on demand without relying on handcrafted rules or global static graphs. By integrating similarity-based retrieval with dynamic dependency validation, DyRetriever enables efficient and adaptive context acquisition. The resulting system, DyCoder, achieves relative Pass@1 improvements of 25.63% and 59.73% on CoderEval and DevEval, respectively, and operates 7.4× faster than static graph baselines.

0 citationsRead paper