Demystifying Gate-Level Localization of RTL Trojans
本文提出一种基于启发式的轻量级方法LoRD，用于检测和定位RTL木马，在合成后的门级网表中通过识别稳定的结构和信号流模式实现高效检测。
0 citationsRead paper
Selected work
Representative Papers
SignRR: Retrieve and Refine Real Motion for Sign Language Production
为解决手语生成中动作连贯性和真实性问题，提出SignRR方法，通过检索真实动作并用残差VQ-VAE优化，以生成连贯且高质量的手语动作序列。
0 citationsRead paper
Should BBR be the default TCP Congestion Control Protocol?
Assessing the feasibility and universality of BBR and its evolutions (BBRv2/v3) as the default TCP congestion control algorithm across heterogeneous network environments. Method: A systematic evaluation combining controlled experiments and comprehensive literature review, comparing BBR variants against mainstream algorithms—including Reno, Cubic, DCTCP, DCQCN, TIMELY, HPCC, and Swift—in diverse settings: the Internet, data centers, Ethernet, wireless networks, and low-Earth-orbit satellite networks. Contribution/Results: BBR achieves significantly higher throughput than Cubic in high-bandwidth, homogeneous-flow scenarios (e.g., 905 Mbps in a gigabit campus network) while maintaining fairness; however, it incurs increased latency and jitter in delay-sensitive deployments, necessitating careful trade-offs. This study is the first to empirically demonstrate that application-level workload characteristics critically influence congestion control protocol selection—providing both theoretical foundations and practical guidance for adaptive, application-aware congestion control deployment.
0 citationsRead paper
GRU-Based Learning for the Identification of Congestion Protocols in TCP Traffic
This study addresses the fine-grained identification of TCP congestion control protocols (Reno, CUBIC, Vegas, BBR) in campus networks. We propose a time-series modeling approach based on Gated Recurrent Units (GRUs), distinguishing itself from conventional CNN- or LSTM-based architectures through its lightweight design—preserving temporal modeling capability while significantly improving inference efficiency for highly dynamic, interference-prone real-world campus networks. Evaluated on real-world, flow-level TCP features—including RTT evolution, congestion window dynamics, and packet-loss patterns—the method achieves 97.04% protocol classification accuracy on unseen traffic, substantially outperforming existing statistical-feature-based and shallow-model approaches. Our key contributions are: (i) empirical validation of lightweight RNNs for protocol identification, demonstrating both efficacy and practical deployability; and (ii) provision of a production-ready deep learning baseline for network measurement and adaptive QoS control.
0 citationsRead paper
Chatbot Deployment Considerations for Application-Agnostic Human-Machine Dialogues
Many enterprises urgently deploy chatbots to enhance customer service efficiency, yet overlook critical prerequisites—including social value alignment and potential societal impacts. This study systematically investigates ethical risks and societal effects of general-purpose human–machine dialogue systems in real-world service contexts, employing computational linguistics and NLP techniques within a multi-case analytical framework. Innovatively, we propose a “Social Values-First” framework that embeds fairness, explainability, and cultural adaptability as core deployment dimensions. Drawing on empirical cases, we derive transferable evaluation principles and implementation pathways. The work bridges the theory–practice gap between AI industrialization and corporate social responsibility, offering both a methodological foundation and actionable guidelines for developing responsible AI systems. (132 words)
0 citationsRead paper
Recent publications
Latest Papers
Demystifying Gate-Level Localization of RTL Trojans
本文提出一种基于启发式的轻量级方法LoRD，用于检测和定位RTL木马，在合成后的门级网表中通过识别稳定的结构和信号流模式实现高效检测。
0 citationsRead paper
SignRR: Retrieve and Refine Real Motion for Sign Language Production
为解决手语生成中动作连贯性和真实性问题，提出SignRR方法，通过检索真实动作并用残差VQ-VAE优化，以生成连贯且高质量的手语动作序列。
0 citationsRead paper
Should BBR be the default TCP Congestion Control Protocol?
Assessing the feasibility and universality of BBR and its evolutions (BBRv2/v3) as the default TCP congestion control algorithm across heterogeneous network environments. Method: A systematic evaluation combining controlled experiments and comprehensive literature review, comparing BBR variants against mainstream algorithms—including Reno, Cubic, DCTCP, DCQCN, TIMELY, HPCC, and Swift—in diverse settings: the Internet, data centers, Ethernet, wireless networks, and low-Earth-orbit satellite networks. Contribution/Results: BBR achieves significantly higher throughput than Cubic in high-bandwidth, homogeneous-flow scenarios (e.g., 905 Mbps in a gigabit campus network) while maintaining fairness; however, it incurs increased latency and jitter in delay-sensitive deployments, necessitating careful trade-offs. This study is the first to empirically demonstrate that application-level workload characteristics critically influence congestion control protocol selection—providing both theoretical foundations and practical guidance for adaptive, application-aware congestion control deployment.
0 citationsRead paper
GRU-Based Learning for the Identification of Congestion Protocols in TCP Traffic
This study addresses the fine-grained identification of TCP congestion control protocols (Reno, CUBIC, Vegas, BBR) in campus networks. We propose a time-series modeling approach based on Gated Recurrent Units (GRUs), distinguishing itself from conventional CNN- or LSTM-based architectures through its lightweight design—preserving temporal modeling capability while significantly improving inference efficiency for highly dynamic, interference-prone real-world campus networks. Evaluated on real-world, flow-level TCP features—including RTT evolution, congestion window dynamics, and packet-loss patterns—the method achieves 97.04% protocol classification accuracy on unseen traffic, substantially outperforming existing statistical-feature-based and shallow-model approaches. Our key contributions are: (i) empirical validation of lightweight RNNs for protocol identification, demonstrating both efficacy and practical deployability; and (ii) provision of a production-ready deep learning baseline for network measurement and adaptive QoS control.
0 citationsRead paper
Chatbot Deployment Considerations for Application-Agnostic Human-Machine Dialogues
Many enterprises urgently deploy chatbots to enhance customer service efficiency, yet overlook critical prerequisites—including social value alignment and potential societal impacts. This study systematically investigates ethical risks and societal effects of general-purpose human–machine dialogue systems in real-world service contexts, employing computational linguistics and NLP techniques within a multi-case analytical framework. Innovatively, we propose a “Social Values-First” framework that embeds fairness, explainability, and cultural adaptability as core deployment dimensions. Drawing on empirical cases, we derive transferable evaluation principles and implementation pathways. The work bridges the theory–practice gap between AI industrialization and corporate social responsibility, offering both a methodological foundation and actionable guidelines for developing responsible AI systems. (132 words)
0 citationsRead paper