Xiaoyu Liu | 刘啸宇

Xiaoyu Liu | 刘啸宇

Huazhong University of Science and Technology

😉 About Me

I am a Master’s student at Huazhong University of Science and Technology, supervised by Prof. Ye Yuan.

Previously, I got my B.Eng degrees (Outstanding Graduates) from Central South University.

My academic research interest includes multi-modal learning, large language model, weakly-supervised learning and vision language pre-training.

My resume can be viewed at Here.

📝 Publications

ArXiv Preprint
sym

PLReMix: Combating Noisy Labels with Pseudo-Label Relaxed Contrastive Representation Learning [PDF] [GitHub]

Xiaoyu Liu, Beitong Zhou, Cheng Cheng

TII 2023
sym

Intelligent fault diagnosis with noisy labels via semi-supervised learning on industrial time series [PDF]

Cheng Cheng, Xiaoyu Liu, Beitong Zhou, Ye Yuan

TCSVT 2024
sym

GTINet: Global Topology-aware Interactions for Unsupervised Point Cloud Registration [PDF]

Yinuo Jiang, Beitong Zhou, Xiaoyu Liu, Qingyi Li, Cheng Cheng

PHM 2024
sym

Surface roughness prediction in boring of TC4-DT with multi-sensor data and machining parameters [Accepted]

Longjianjie Zhang, Xiaoyu Liu, Ziheng Zhou, Zhihui Wang, Feng Hua, Cheng Cheng

💼 Internships

Baidu, Inc.Jan. 2024 - Now

Large Language Model Algorithm Intern

  • Responsible for the content production pipeline of Baidu Zhidao, collecting relevant materials from the entire internet based on user query data, fine-tuning LLMs to generate answers, and deploying a low-cost content production line with a daily output of hundreds of thousands.

  • Develop a multimodal agent routing-based response generation scheme to tackle issues arising from crucial information concealed in images. This enables the model to autonomously choose a response generation scheme based on plain text/+OCR/+original images according to the amount of information and relevance, achieving a 29\% improvement in content usability.

💡 Projects

Blade Knife Automatic Selection with LLMJan. 2024 - Now

Project Leader

  • In the process of aero-engine blade machining, use CatBoost to predict a reasonable initial value of the knife size, then use LLM to automate the tool screening process until it passes the system simulation test, avoiding the tedious steps of manual trial and error.

  • Design and develop a natural language interactive agent-based machining process control system, which operates serialized process objects through an LLM and realizes the fine adjustment of process flow and parameters.

  • Cooperate with JITRI - Jiangsu Industrial Technology Research Institute.

LLM Agent Anime Characters ChatbotNov. 2023 - Dec. 2023

Independent Developer

  • Build an LLM chatbot based on ModelScope, LangChain and Gradio, use prompts and document retrieval to give characters different personalities and memories. Demo

  • Won the best popularity award in Tongyi Qianwen Agent Builder Creative Challenge Track. (20/654)

Intelligent Anti-epidemic Disinfection VehicleSep. 2020 - May 2021

Project Leader

  • Develop an SSD-based mask-wearing detection algorithm for the anti-epidemic disinfection vehicle, which is deployed on Raspberry Pi 4b, with recognition accuracy reaching 89% and the speed up to 5 f/s.

  • Develop the website backend based on Flask, which controls and monitors the vehicle through MQTT and RTMP.

  • Collect one first-class national prize and two third-class national prizes.

MIT-BIH ECG Arrhythmia DiagnosisMar. 2020 - June 2022

Project Leader

  • Propose an ECG arrhythmia diagnosis algorithm based on time-series Transformer auto-encoder and GWO-SVM Optimizer.

  • Develop a knowledge graph-based consultation system for cardiovascular diseases with Scrapy crawling the data and Neo4j building the graph database. Code

  • Be selected as the excellent project in the National Undergraduate Innovation and Entrepreneurship Project by Central South University. (10%)

📖 Educations

Huazhong University of Science and TechnologySep. 2022 - June 2025 (expected)

  • M.Sc. in Control Science and EngineeringGrade: 91.35 / 100

Central South UniversitySep. 2018 - June 2022

  • B.E. in Intelligent Science and TechnologyGrade: 90.91 / 100

🎖 Honors and Awards

First-class Scholarship for Postgraduates - Huazhong University of Science and Technology2022

Outstanding Graduates - Central South University2022

Excellent Student (10%) - Central South University2019/2020/2021

First-class National Prize (Top-5 in China) - National College Student Software Innovation Competition2020

Third-class National Prize (15% in China) - China Robotics and Artificial Intelligence Competition2020

Third-class National Prize (20% in China) - Undergraduate Electronic Design Contest - 2020 Embedded System Design Invitational Contest2020

Third-class National Prize (Top-20 in China) - Chinese Collegiate Computing Competition2021

💻 Open Source