Zeju Li 李泽钜

Zeju Li 李泽钜

PhD @ CUHK

Internships

  • 2021.06-2022.12 State Key Laboratory of Networking And Switching Technology

    • Interests: NLP and Knowledge graph
    • Duty: I constructed more than one hundred knowledge graphs of Electronic Information automatically by NER and RegEx; Our team built a web system that can mark examination papers automatically by OCR and vue, and we applied a patent for it; Our team studied KG fusion and entity alignment, and we propose an automated knowledge extraction, data cleaning and graph fusion framework for heterogeneous learning resources. Our paper got best paper nomination award of ``ICALT2023”.

  • 2022.05-2023.5 National Engineering Research Center of Mobile Network Security

    • Interest: Multi-modal and Semantic communication (Mentor: Prof. Guoshun Nan)
    • Duty: I do research in AI for Communication and Semantic communication based on deep learning. We propose a practical physical layer black-box attack method to destroy the robustness of deep learning-based semantic communication systems, where the adversaries are imperceptible to humans. And then, we propose SemProtector: A Unified Framework for Semantic Protection in Deep Learning-based Semantic Communication Systems. Our works were published by VTC2022-fall, ICC2023 and IEEE Communications Magazine.

  • 2023.04-2023.07 The National Laboratory of Pattern Recognition (NLPR) at the Institute of Automation of the Chinese Academy of Sciences (CASIA).

    • Interest: Diffusion model and AI4Science.
    • Duty: I do research in Diffusion model and AI4Science. I study VAE, DDPM and DDIM, and try to solve combinatorial optimization problems by DDPM.

  • 2023.08-Now Beijing Shuyuan Digital City Research Center

    • Interest: 3D Multimodal LLM and scene understanding.
    • Duty: I do research in 3D-LLM and 3d scene understanding. We introduce a novel and efficient prompt tuning paradigm, 3DMIT. This paradigm eliminates the alignment stage between 3D scenes and language and extends the instruction prompt with the 3D modality information including the entire scene and segmented objects. Our code is available at https://github.com/staymylove/3DMIT.

  • 2023.07-2024.04 GAP Lab, The Chinese University of Hong Kong (Shenzhen)

    • Interest: 3d generation and 3d model edit. (supervised by Xiaoguang Han)
    • Duty: I do research in 3d generation (sketch to 3D) and 3d model edit. We work on the topic ’Sketch to 3D’, using text prompts and sketch to controllable generate high-quality 3D objects. Besides, I work on the topic of human reconstruction with gaussian splatting. I finish my bachelor thesis in GAP lab.

  • 2024.04-Now King Abdullah University of Science and Technology (KAUST)

    • Interest: LLM continual learning (supervised by Mohamed Elhoseiny)
    • Duty: I do research in LLM continual learning, transfer learning as a visiting student.

  • 2024.05-Now Cure Lab, The Chinese University of Hong Kong

    • Interest: Code-LLM, tokenizer, RTL LLM alignment. (supervised by Qiang Xu)
    • Duty: I do research in RTL(verilog) represent learning and understanding by LLM, alignment of HLS and RTL, Large Circuit Model.

  • 2024.05-Now Nanjing Industrial Innovation Center of EDA

    • Interest: Code-LLM, tokenizer, RTL LLM alignment. (supervised by Qiang Xu)
    • Duty: I do research in RTL(verilog) represent learning and understanding by LLM, alignment of HLS and RTL, Large Circuit Model.