Guangyao Chen

CGY.png

I received my Ph.D. degree at National Engineering Research Center of Visual Technology, School of Computer Science, Peking University, supervised by Prof.Yonghong Tian. I received my Bachelor degree in Computer Science from Wuhan University in 2018.

My research vision is to devise an efficient, reliable, and autonomous AI system that can operate in the open world. Guided by the principles of efficiency, reliability, and autonomy, I concentrate on these crucial aspects to enhance and adapt AI models:

  • Autonomous Agent: constructing efficient and robust autonomous LLM-based agent systems that leverage LLMs to accomplish automated planning, reasoning, learning, and collaboration.
  • Open World Learning: boosting the robustness of efficient AI models in the open world, achieving adaptation to diverse distributions, and discovering and learning new categories.
  • Efficient AI Model: designing an efficient AI model architecture and application criteria, aiming to empower AI models to run on resource-scarce devices, and facilitate efficient inference and learning.

Feel free to catch me if interested to discuss ideas or work together. 😜

news

2024.04  🎉🎉 One paper on Autonomous Agent is accepted by the conference IJCAI 2024.
2023.12  🎉🎉 One paper on Incremental Novel Class Discovery is accepted by the conference AAAI 2024.
2023.11 Invited talk at Qingyuan Workshop (Online).
2023.11 Invited talk at RLChina 2023 in Suzhou.
2023.06 I receive my Ph.D. degree in computer science from Peking University with Outstanding Doctoral Dissertation Award.
2022.09  🎉🎉 Two papers on Image-based Reinforcement Learning and Out-of-Distribution Detection are accepted by the conference NeurIPS 2022.

selected publications

  1. ECCV
    Learning open set network with discriminative reciprocal points
    Guangyao Chen, Limeng Qiao, Yemin Shi, and 5 more authors
    In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part III 16, 2020 (Spotlight Presentation, Acceptance Rate: < 5%)
    arXiv Paper Dataset Code
  2. ICCV
    Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain
    Guangyao Chen, Peixi Peng, Li Ma, and 3 more authors
    In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Oct 2021
    arXiv Paper Code
  3. TPAMI
    Adversarial Reciprocal Points Learning for Open Set Recognition
    Guangyao Chen, Peixi Peng, Xiangqian Wang, and 1 more author
    IEEE Transactions on Pattern Analysis and Machine Intelligence, Oct 2022 (Academician Shi Qingyun’s Outstanding Paper Award)
    arXiv Paper Code
  4. ACM MM
    Annotation-Efficient Untrimmed Video Action Recognition
    Yixiong Zou, Shanghang Zhang, Guangyao Chen, and 3 more authors
    In Proceedings of the 29th ACM International Conference on Multimedia, Oct 2021
    arXiv Paper
  5. NeurIPS
    Spectrum Random Masking for Generalization in Image-based Reinforcement Learning
    Yangru Huang, Peixi Peng, Yifan Zhao, and 2 more authors
    In Advances in Neural Information Processing Systems, Oct 2022
    Paper
  6. NeurIPS
    OpenOOD: Benchmarking Generalized Out-of-Distribution Detection
    Jingkang Yang, Pengyun Wang, Dejian Zou, and 13 more authors
    In Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track, Oct 2022
    arXiv Paper Code
  7. Preprint
    Training Full Spike Neural Networks via Auxiliary Accumulation Pathway
    Guangyao Chen, Peixi Peng, Guoqi Li, and 1 more author
    arXiv, Oct 2023
    arXiv
  8. IJCAI
    AutoAgents: A Framework for Automatic Agent Generation
    Guangyao Chen*, Siwei Dong*, Yu Shu*, and 5 more authors
    In The 33rd International Joint Conference on Artificial Intelligence (IJCAI-24), Oct 2024
    arXiv Code
  9. AAAI
    Adaptive Discovering and Merging for Incremental Novel Class Discovery
    Guangyao Chen, Peixi Peng, Yangru Huang, and 2 more authors
    In Thirty-Eighth AAAI Conference on Artificial Intelligence, Oct 2024
    arXiv Paper