Selected work

Representative Papers

Data-Driven Radio Propagation Modeling using Graph Neural Networks

Jan 08, 2025

To address the low accuracy and high computational cost of radio signal propagation modeling in real-world scenarios, this paper proposes a data-driven coverage map generation method based on Graph Neural Networks (GNNs). The approach automatically constructs a heterogeneous graph from environmental imagery, jointly encoding spatial topology and ray-based propagation relationships, and learns propagation patterns end-to-end using sparse, real-world signal measurements as supervision. This work is the first to apply GNNs to empirical radio propagation modeling, eliminating reliance on traditional physics-based solvers. Experiments demonstrate that the method surpasses classical numerical solvers (e.g., FDTD) and heuristic models in coverage map reconstruction accuracy, achieves 10–100× faster inference, and exhibits strong generalization—enabling high-fidelity signal coverage prediction from only a few measurements. These advances significantly enhance the efficiency and practicality of wireless network deployment.

2 citationsRead paper

Knowledge Distillation for LLM-Based Human Activity Recognition in Homes

Jan 12, 2026

This study addresses the high computational cost of deploying large language models (LLMs) for human activity recognition (HAR) in home environments. It presents the first systematic evaluation of LLMs of varying scales on HAR tasks and introduces a knowledge distillation–based model compression approach. By transferring the reasoning capabilities of a large LLM to a lightweight student model, the method achieves comparable accuracy to the best-performing large model while retaining only 1/50th of its original parameters. Experimental results on two mainstream HAR datasets demonstrate that the distilled compact model maintains near-optimal performance, offering an efficient pathway for deploying LLM-driven HAR systems in resource-constrained settings.

1 citationsRead paper

End-to-end Early Classification of Time Series in Non-Stationary Environments

Aug 20, 2026

研究在非平稳环境下时间序列的早期分类问题，通过强化学习提出DQeND模型联合学习表示、分类和触发决策，优于传统分离设计方法。

0 citationsRead paper

A Study of Kernel Telemetry Options for Security-Oriented Provenance

Aug 11, 2026

Existing provenance tracing systems struggle to simultaneously ensure security, integrity, and availability in production environments. This study presents the first systematic classification and evaluation of eight provenance systems and five capture agents, with a focus on the heterogeneity and security weaknesses of their underlying kernel telemetry mechanisms. Leveraging eBPF for fine-grained event filtering—such as container-level monitoring—and employing microbenchmarks to analyze performance overhead, the work reveals that most tools fail to guarantee the integrity and availability of event data, rendering them inadequate for security-critical scenarios. The findings provide an empirical foundation and design guidance for building high-fidelity, low-overhead provenance capture systems suitable for real-world deployment.

0 citationsRead paper

Tracing sources of epistemic uncertainty in deep learning predictions: homo- and hetero-scedastic linearized estimators

Aug 07, 2026

This work addresses the challenge of disentangling epistemic uncertainty from uncertainty induced by local data sparsity in deep learning, which are often conflated and difficult to distinguish. The authors integrate two classical statistical uncertainty estimation techniques into a deep learning framework, constructing homoscedastic and heteroscedastic linearized estimators based on an approximate Fisher information matrix. This approach enables, for the first time, a scalable and fine-grained decomposition of epistemic uncertainty and the effects of local data scarcity in modern deep neural networks. Experimental results demonstrate that the proposed method effectively quantifies the relative influence of each uncertainty type across individual test samples, substantially enhancing model robustness and reliability in real-world applications.

0 citationsRead paper

Recent publications

Latest Papers

End-to-end Early Classification of Time Series in Non-Stationary Environments

Aug 20, 2026

研究在非平稳环境下时间序列的早期分类问题，通过强化学习提出DQeND模型联合学习表示、分类和触发决策，优于传统分离设计方法。

0 citationsRead paper

A Study of Kernel Telemetry Options for Security-Oriented Provenance

Aug 11, 2026

Existing provenance tracing systems struggle to simultaneously ensure security, integrity, and availability in production environments. This study presents the first systematic classification and evaluation of eight provenance systems and five capture agents, with a focus on the heterogeneity and security weaknesses of their underlying kernel telemetry mechanisms. Leveraging eBPF for fine-grained event filtering—such as container-level monitoring—and employing microbenchmarks to analyze performance overhead, the work reveals that most tools fail to guarantee the integrity and availability of event data, rendering them inadequate for security-critical scenarios. The findings provide an empirical foundation and design guidance for building high-fidelity, low-overhead provenance capture systems suitable for real-world deployment.

0 citationsRead paper

Tracing sources of epistemic uncertainty in deep learning predictions: homo- and hetero-scedastic linearized estimators

Aug 07, 2026

This work addresses the challenge of disentangling epistemic uncertainty from uncertainty induced by local data sparsity in deep learning, which are often conflated and difficult to distinguish. The authors integrate two classical statistical uncertainty estimation techniques into a deep learning framework, constructing homoscedastic and heteroscedastic linearized estimators based on an approximate Fisher information matrix. This approach enables, for the first time, a scalable and fine-grained decomposition of epistemic uncertainty and the effects of local data scarcity in modern deep neural networks. Experimental results demonstrate that the proposed method effectively quantifies the relative influence of each uncertainty type across individual test samples, substantially enhancing model robustness and reliability in real-world applications.

0 citationsRead paper

How Far Do Simple Transformations Translate Across Text Embedding Models?

Aug 06, 2026

This study investigates whether latent representations from heterogeneous text embedding models can be transferred via simple transformations to enable direct AI-to-AI communication without decoding into human-readable text. For the first time, we systematically evaluate the effectiveness and limitations of linear mappings as lightweight translators across nine diverse models varying in architecture, pooling strategy, and training objective, using real-world textual data. Through comprehensive metrics—including Centered Kernel Alignment (CKA) similarity, downstream task transfer performance, fidelity, and retrieval accuracy—we find that simple transformations succeed only between partially compatible model pairs and largely fail otherwise. These results indicate that semantic transfer across heterogeneous embedding spaces cannot be universally achieved through alignment alone, as compatibility is jointly constrained by architectural design, training objectives, pooling mechanisms, and data distribution.

0 citationsRead paper

Learning Compression Rules for Network Traffic

Aug 05, 2026

This work addresses the high redundancy in header fields of structured network traffic by proposing RECAP, an automated rule compression method that operates without human intervention. RECAP innovatively integrates unsupervised structure discovery with constrained optimization: it first recursively partitions packets using a normalized entropy ratio criterion to identify redundant patterns, then selects the subset of rules yielding maximal compression gain under a hard constraint on rule count via dynamic programming, and finally applies short-code replacements in accordance with the IETF SCHC standard. Experiments on four real-world datasets from IoT and 5G core networks demonstrate that RECAP significantly outperforms expert-designed compression rule sets, achieving substantially higher compression efficiency with only a small number of learned rules.

0 citationsRead paper