Selected work

Representative Papers

CoeF-SFL: Preserving Collaborative Server-Client Learning with Enhanced Communication Efficiency

Sep 28, 2026

This study addresses the high communication overhead and optimization objective misalignment caused by auxiliary networks in split federated learning by proposing the CoeF-SFL framework. This method substantially reduces communication costs through single data exchange per round and gradient reuse, while eliminating auxiliary networks to preserve end-to-end optimization objectives. Furthermore, it introduces a curvature-based gradient compensation mechanism that leverages diagonal Hessian approximation and Jacobian-Hessian surrogate losses to correct stale gradients within the activation space. Experimental results demonstrate that the proposed framework significantly outperforms existing methods across both vision and language tasks. The source code has been made publicly available.

0 citationsRead paper

When the Model Retires: An Empirical Study of LLM Migration in Open-Source Applications

Sep 25, 2026

This study addresses the service disruptions and delayed migrations in open-source applications caused by the deprecation of commercial large language models. Leveraging GitHub data, this work conducts an empirical investigation using commit mining, manual annotation, statistical reweighting, and dependency modeling to provide the first quantitative analysis of developer migration behaviors and the impact mechanisms of notification policies. The findings reveal that 82% of migrations occur only after service failures, highlighting a significant correlation between notification duration and migration timeliness, as well as the widespread prevalence of hardcoding practices. Furthermore, this research releases an associated dataset, offering empirical evidence to inform model deprecation strategies and optimize development toolchains for more resilient software ecosystems.

0 citationsRead paper

What to Keep and What to Drop: Adaptive Table Filtering Framework

Jun 29, 2025

Large language models (LLMs) suffer from degraded table reasoning performance due to input-length constraints, hindering effective processing of long, wide tables. Method: We propose Adaptive Table Filtering (ATF), a plug-and-play framework that requires no model fine-tuning. ATF dynamically identifies and retains query-relevant table regions via question-aware column semantic description generation, hierarchical clustering, and sparse–dense vector alignment scoring. Its modular design enables cross-task adaptive balancing between information preservation and structural simplification. Contribution/Results: ATF prunes ~70% of table cells on average, significantly improving reasoning accuracy across diverse TableQA benchmarks. Only in rare cases requiring full-table structural understanding does performance marginally decline. Crucially, ATF is the first approach to jointly model LLM-driven semantic comprehension with interpretable, structured filtering—achieving strong efficiency, task-agnostic generalizability, and parameter-free transferability.

0 citationsRead paper

ReCDAP: Relation-Based Conditional Diffusion with Attention Pooling for Few-Shot Knowledge Graph Completion

May 12, 2025

To address the degradation of few-shot link prediction performance caused by long-tail relations in knowledge graphs, this paper proposes Relation-Conditioned Diffusion with Attention Pooling (ReCDAP). ReCDAP introduces negative triples as structured signals into diffusion modeling for the first time, establishing a dual-path latent distribution framework that separately models positive and negative relations. It further employs a relation-aware attention pooling mechanism to explicitly capture discriminative differences between them. Integrated with few-shot meta-learning and negative sampling augmentation, ReCDAP jointly models both semantic and discriminative information of triples under sparse relations. Extensive experiments on FB15k-237 and NELL-995 demonstrate that ReCDAP significantly improves link prediction accuracy in few-shot settings, achieving state-of-the-art (SOTA) performance.

0 citationsRead paper

Recent publications

Latest Papers

CoeF-SFL: Preserving Collaborative Server-Client Learning with Enhanced Communication Efficiency

Sep 28, 2026

This study addresses the high communication overhead and optimization objective misalignment caused by auxiliary networks in split federated learning by proposing the CoeF-SFL framework. This method substantially reduces communication costs through single data exchange per round and gradient reuse, while eliminating auxiliary networks to preserve end-to-end optimization objectives. Furthermore, it introduces a curvature-based gradient compensation mechanism that leverages diagonal Hessian approximation and Jacobian-Hessian surrogate losses to correct stale gradients within the activation space. Experimental results demonstrate that the proposed framework significantly outperforms existing methods across both vision and language tasks. The source code has been made publicly available.

0 citationsRead paper

When the Model Retires: An Empirical Study of LLM Migration in Open-Source Applications

Sep 25, 2026

This study addresses the service disruptions and delayed migrations in open-source applications caused by the deprecation of commercial large language models. Leveraging GitHub data, this work conducts an empirical investigation using commit mining, manual annotation, statistical reweighting, and dependency modeling to provide the first quantitative analysis of developer migration behaviors and the impact mechanisms of notification policies. The findings reveal that 82% of migrations occur only after service failures, highlighting a significant correlation between notification duration and migration timeliness, as well as the widespread prevalence of hardcoding practices. Furthermore, this research releases an associated dataset, offering empirical evidence to inform model deprecation strategies and optimize development toolchains for more resilient software ecosystems.

0 citationsRead paper

What to Keep and What to Drop: Adaptive Table Filtering Framework

Jun 29, 2025

Large language models (LLMs) suffer from degraded table reasoning performance due to input-length constraints, hindering effective processing of long, wide tables. Method: We propose Adaptive Table Filtering (ATF), a plug-and-play framework that requires no model fine-tuning. ATF dynamically identifies and retains query-relevant table regions via question-aware column semantic description generation, hierarchical clustering, and sparse–dense vector alignment scoring. Its modular design enables cross-task adaptive balancing between information preservation and structural simplification. Contribution/Results: ATF prunes ~70% of table cells on average, significantly improving reasoning accuracy across diverse TableQA benchmarks. Only in rare cases requiring full-table structural understanding does performance marginally decline. Crucially, ATF is the first approach to jointly model LLM-driven semantic comprehension with interpretable, structured filtering—achieving strong efficiency, task-agnostic generalizability, and parameter-free transferability.

0 citationsRead paper

ReCDAP: Relation-Based Conditional Diffusion with Attention Pooling for Few-Shot Knowledge Graph Completion

May 12, 2025

To address the degradation of few-shot link prediction performance caused by long-tail relations in knowledge graphs, this paper proposes Relation-Conditioned Diffusion with Attention Pooling (ReCDAP). ReCDAP introduces negative triples as structured signals into diffusion modeling for the first time, establishing a dual-path latent distribution framework that separately models positive and negative relations. It further employs a relation-aware attention pooling mechanism to explicitly capture discriminative differences between them. Integrated with few-shot meta-learning and negative sampling augmentation, ReCDAP jointly models both semantic and discriminative information of triples under sparse relations. Extensive experiments on FB15k-237 and NELL-995 demonstrate that ReCDAP significantly improves link prediction accuracy in few-shot settings, achieving state-of-the-art (SOTA) performance.

0 citationsRead paper