About

Songwu Lu is a Professor in the Department of Computer Science at the University of California, Los Angeles. He leads the Wireless Networking Group (WiNG) at UCLA, focusing his research on wireless networking, mobile systems, cloud computing, and wireless and Internet security. Prior to his tenure at UCLA, he earned his Ph.D. from the University of Illinois at Urbana-Champaign in 1999. His research encompasses a broad range of topics including ad hoc network security, data dissemination protocols for sensor networks, rate adaptation in wireless networks, and security in mobile ad hoc networks. He has contributed to the field through numerous publications, including book chapters, journal articles, and conference papers, addressing critical issues in wireless communication, network security, and system robustness.

Research topics

• Computer Science
• Computer network
• Telecommunications
• Computer Security
• Operating system
• Data science
• Engineering

Selected publications

• SibylSense: Adaptive Rubric Learning via Memory Tuning and Adversarial Probing

  arXiv (Cornell University) · 2026-02-24

  articleOpen access

  Designing aligned and robust rewards for open-ended generation remains a key barrier to RL post-training. Rubrics provide structured, interpretable supervision, but scaling rubric construction is difficult: expert rubrics are costly, prompted rubrics are often superficial or inconsistent, and fixed-pool discriminative rubrics can saturate and drift, enabling reward hacking. We present SibylSense, an inference-time learning approach that adapts a frozen rubric generator through a tunable memory bank of validated rubric items. Memory is updated via verifier-based item rewards measured by reference-candidate answer discriminative gaps from a handful of examples. SibylSense alternates memory tuning with a rubric-adversarial policy update that produces rubric-satisfying candidate answers, shrinking discriminative gaps and driving the rubric generator to capture new quality dimensions. Experiments on two open-ended tasks show that SibylSense yields more discriminative rubrics and improves downstream RL performance over static and non-adaptive baselines.

  PublisherOA PDF

• SibylSense: Adaptive Rubric Learning via Memory Tuning and Adversarial Probing

  Open MIND · 2026-02-24

  preprint

  Designing aligned and robust rewards for open-ended generation remains a key barrier to RL post-training. Rubrics provide structured, interpretable supervision, but scaling rubric construction is difficult: expert rubrics are costly, prompted rubrics are often superficial or inconsistent, and fixed-pool discriminative rubrics can saturate and drift, enabling reward hacking. We present SibylSense, an inference-time learning approach that adapts a frozen rubric generator through a tunable memory bank of validated rubric items. Memory is updated via verifier-based item rewards measured by reference-candidate answer discriminative gaps from a handful of examples. SibylSense alternates memory tuning with a rubric-adversarial policy update that produces rubric-satisfying candidate answers, shrinking discriminative gaps and driving the rubric generator to capture new quality dimensions. Experiments on two open-ended tasks show that SibylSense yields more discriminative rubrics and improves downstream RL performance over static and non-adaptive baselines.

  DOI

• AnyPro: Preference-Preserving Anycast Optimization based on Strategic AS-Path Prepending

  arXiv (Cornell University) · 2026-03-22

  articleOpen access

  Operating large-scale anycast networks is challenging because client-to-site mappings often misalign with operator's expectation due to opaque inter-domain routing. We present AnyPro, the first system to unlock the full potential of AS-path prepending (ASPP), efficiently deriving globally optimal configurations to steer clients toward performance-optimal sites at scale. AnyPro first employs an efficient polling mechanism to identify all clients sensitive to ASPP. By analyzing the routing changes during the process, the system derives a set of ASPP constraints that guide client traffic toward the desired sites. We then formulate the anycast optimization problem as a constraint-based program and compute optimal ASPP configurations. Extensive evaluation on a global testbed with 20 PoPs demonstrates the effectiveness of AnyPro: it reduces the 90th percentile latency by 37.7% compared to baseline configurations without ASPP. Furthermore, we show that AnyPro can be integrated with PoP-level anycast optimization techniques to achieve additional performance gains.

  PublisherOA PDF

• Robustness Analysis of Mega-constellation Networks Against Node Attacks

  2025-04-11 · 1 citations

  article

  Mega-constellation networks have received increasing attention in recent years. Network robustness is important for mega-constellation networks, but it is also difficult because satellites are expensive to maintain and reconfigure in outer space. This paper presents some network robustness evaluations on mega-constellation networks. Firstly, a definition of the importance metric of satellites is proposed based on traffic due to the difference between constellation networks and terrestrial networks. Based on this, three node attack strategies are defined to mimic actual attack behavior against mega-constellation networks. Simulations are implemented in three typical mega-constellation network models and the robustness is evaluated using a newly presented metric. The results show that the random attack strategy has a wider impact on mega-constellation networks, and the selective attacks may have a stronger impact on the network, which results in a greater packet loss ratio in mega-constellations.

  PublisherDOI

• Analysis of intermittent distributed generation capacity of multi load access honeycomb distribution network

  Measurement · 2025-12-19

  article
  PublisherDOI

• Optimization of Smartphone-Based Strain Measurement Algorithm Utilizing Arc-Support Line Segments

  Buildings · 2025-09-20

  articleOpen access

  Smartphone-based strain monitoring of structural components is an emerging approach to structural health monitoring. However, the existing techniques suffer from limited accuracy and poor cross-device adaptability. This study aims to optimize the smartphone-based Micro Image Strain Sensing (MISS) method by replacing the traditional Connected Component Labeling (CCL) algorithm with the arc-support line segments (ASLS) algorithm, thereby significantly enhancing the stability and adaptability of circle detection in micro-images captured by diverse smartphones. Additionally, this study evaluates the impact of lighting conditions and lens distortion on the optimized MISS method. The experimental results demonstrate that the ASLS algorithm outperforms CCL in terms of recognition accuracy (maximum error of 0.94%) and cross-device adaptability, exhibiting greater robustness against color temperature and focal length variations. Under fluctuating lighting conditions, the strain measurement noise remains within ±0.5 με and with a maximum error of 7.0 με compared to LVDT measurements, indicating the strong adaptability of the optimized MISS method to external light changes. Barrel distortion in microscopic images induces a maximum pixel error of 5.66%, yet the final optimized MISS method achieves highly accurate strain measurements. The optimized MISS method significantly improves measurement stability and engineering applicability, enabling effective large-scale implementation for strain monitoring of civil infrastructure.

  PublisherOA PDFDOI

• RLTHF: Targeted Human Feedback for LLM Alignment

  ArXiv.org · 2025-02-19

  preprintOpen access

  Fine-tuning large language models (LLMs) to align with user preferences is challenging due to the high cost of quality human annotations in Reinforcement Learning from Human Feedback (RLHF) and the generalizability limitations of AI Feedback. To address these challenges, we propose RLTHF, a human-AI hybrid framework that combines LLM-based initial alignment with selective human annotations to achieve full-human annotation alignment with minimal effort. RLTHF identifies hard-to-annotate samples mislabeled by LLMs using a reward model's reward distribution and iteratively enhances alignment by integrating strategic human corrections while leveraging LLM's correctly labeled samples. Evaluations on HH-RLHF and TL;DR datasets show that RLTHF reaches full-human annotation-level alignment with only 6-7% of the human annotation effort. Furthermore, models trained on RLTHF's curated datasets for downstream tasks outperform those trained on fully human-annotated datasets, underscoring the effectiveness of RLTHF.

  PublisherOA PDFDOI

• Direct Reasoning Optimization: Token-Level Reasoning Reflectivity Meets Rubric Gates for Unverifiable Tasks

  ArXiv.org · 2025-06-16

  preprintOpen access

  Reinforcement learning (RL) training of large language models (LLMs) on unverifiable tasks is challenging even when a reasonable-quality reference answer is available. We propose a constrained RL training framework that (i) optimizes a token-level dense Reasoning Reflection Reward (R3) aligned with reasoning quality, and (ii) enforces rubric-gating as feasibility constraints at the rollout group level. R3 measures the model's token-level certainty of a reference answer under its chain-of-thought (CoT) prefix, and selectively emphasizes tokens with high cross-rollout variance, which we call reasoning-reflective tokens, that would otherwise be diluted by the bulk of low-variance tokens. The same variance signal also drives a filter that discards queries with insufficient signal for comparative learning. Rubric-gating complements R3 by operationalizing principled task criteria as hard accept/reject checks on final answers. Empirically, across four datasets spanning scientific writing, medicine, legal contracts, and finance, our framework outperforms strong baselines, achieves faster, more sample-efficient learning, and respects feasibility constraints.

  PublisherOA PDFDOI

• Taming the Insecurity of Cellular Emergency Services (9–1-1): From Vulnerabilities to Secure Designs

  IEEE/ACM Transactions on Networking · 2024-03-26

  articleSenior author

  Cellular networks, vital for delivering emergency services, enable mobile users to dial emergency calls (e.g., 9–1-1 in the U.S.), which are forwarded to public safety answer points (PSAPs). Regulatory requirements allow anonymous user equipment (UE) without a SIM card or valid mobile subscription to access these services. However, supporting emergency services for anonymous UEs introduces different operations, expanding the attack surface of cellular infrastructure. In this study, we explore the insecurity of cellular emergency services, identifying six security vulnerabilities. These vulnerabilities can be exploited for free data service attacks against carriers and data DoS/overcharge and denial of cellular emergency service (DoCES) attacks against mobile users. Experimental validation in networks of three major U.S. carriers and two major Taiwan carriers demonstrates the global impact of our findings. Finally, we propose and prototype standard-compliant remedies to mitigate these vulnerabilities.

  PublisherDOI

• CloudEval-YAML: A Practical Benchmark for Cloud Configuration Generation

  arXiv (Cornell University) · 2023-11-10 · 4 citations

  preprintOpen access

  Among the thriving ecosystem of cloud computing and the proliferation of Large Language Model (LLM)-based code generation tools, there is a lack of benchmarking for code generation in cloud-native applications. In response to this need, we present CloudEval-YAML, a practical benchmark for cloud configuration generation. CloudEval-YAML tackles the diversity challenge by focusing on YAML, the de facto standard of numerous cloud-native tools. We develop the CloudEval-YAML benchmark with practicality in mind: the dataset consists of hand-written problems with unit tests targeting practical scenarios. We further enhanced the dataset to meet practical needs by rephrasing questions in a concise, abbreviated, and bilingual manner. The dataset consists of 1011 problems that take more than 1200 human hours to complete. To improve practicality during evaluation, we build a scalable evaluation platform for CloudEval-YAML that achieves a 20 times speedup over a single machine. To the best of our knowledge, the CloudEval-YAML dataset is the first hand-written dataset targeting cloud-native applications. We present an in-depth evaluation of 12 LLMs, leading to a deeper understanding of the problems and LLMs, as well as effective methods to improve task performance and reduce cost.

  PublisherOA PDFDOI

Recent grants

• TWC: Small: Collaborative: Cellular Network Services in Peril: A Perspective on Control-Plane and Data-Plane Design

  NSF · $261k · 2015–2018

• NeTS: Small: Collaborative Research: Identifying Design and Operational Loopholes in Cellular Network Protocols

  NSF · $251k · 2014–2018

• CNS Core: Small: A Software-Defined Approach to 5G Services for Robots

  NSF · $400k · 2019–2023

• TWC: Small: Collaborative: Secure Data Charging Architecture for Mobile Devices in 3G/4G Cellular Networks: Vulnerabilities and Solutions

  NSF · $252k · 2014–2018

• CSR: Small: A Server-Centric Approach to Data Center Networks

  NSF · $497k · 2010–2014

Frequent coauthors

• Haiyun Luo

  Tsinghua University

  35 shared

• Chunyi Peng

  34 shared

• Yuanjie Li

  30 shared

• Lixia Zhang

  UCLA Health

  27 shared

• Chi-Yu Li

  National Yang Ming Chiao Tung University

  24 shared

• Zhaowei Tan

  Guangdong Academy of Sciences

  24 shared

• Xinbing Wang

  Chinese Academy of Agricultural Sciences

  20 shared

• Guan-Hua Tu

  Michigan State University

  19 shared

Education

• Ph.D.

  University of Illinois at Urbana-Champaign

  1999