Jaeyun Song

Ph.D student at Graduate School of AI, Korea Advanced Institute of Science and Technology (KAIST).

I’m a Ph.D student at Machine Learning and Intelligence Lab (MLILAB) in KAIST, under the supervision of Prof. Eunho Yang.

My research goal is to enhance privacy preservation and robustness in the training and inference of large generative models. Initially, I focused on addressing robustness challenges, such as class imbalance (ICLR’22, ICML’22), structural noise in GNNs (ICML’23), and the dataset bias (NeurIPS’24). Recently, my work has shifted toward advancing privacy-preserving machine learning, where I proposed an efficient private inference method for large diffusion models. My current research aims to develop an efficient private inference method for large language models.

Education

Conference Publications

  • A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective
    Yeonsung Jung*, Jaeyun Song*, June Yong Yang, Jin-Hwa Kim, Sung-Yub Kim, Eunho Yang (*: equal contribution)
    NeurIPS 2024

  • RGE: A Repulsive Graph Rectification for Node Classification via Influence
    [paper] [code]
    Jaeyun Song*, SungYub Kim*, Eunho Yang (*: equal contribution)
    ICML 2023

  • WeavSpeech: Data Augmentation Strategy for Automatic Speech Recognition via Semantic-Aware Weaving
    [paper]
    Kyusung Seo, Joonhyung Park, Jaeyun Song, Eunho Yang
    ICASSP 2023

  • TAM: Topology-Aware Margin Loss for Class-Imbalanced Node Classification
    [paper] [code]
    Jaeyun Song*, Joonhyung Park*, Eunho Yang (*: equal contribution)
    ICML 2022

  • GraphENS: Neighbor-Aware Ego Network Synthesis for Class-Imbalanced Node Classification
    [paper] [code]
    Joonhyung Park*, Jaeyun Song*, Eunho Yang (*: equal contribution)
    ICLR 2022

  • Learning How Long to Wait: Adaptively-Constrained Monotonic Multihead Attention for Streaming ASR
    [paper]
    Jaeyun Song, Hajin Shim, Eunho Yang
    ASRU 2021

  • Mutually-Constrained Monotonic Multihead Attention for Online ASR
    [paper]
    Jaeyun Song, Hajin Shim, Eunho Yang
    ICASSP 2021

Work Experiences

  • Intern, SK Hynix, Icheon, South Korea (Dec 2018 - Feb 2019)

Projects

  • Development of Technology for Content Creation and Automated Correction Based on Deep Learning, Naver AI, 2021-2023

  • Development of brain-inspired AI with human-like intelligence, IITP, 2019-2022

Academic Services

  • Conference Reviewer: NeurIPS 2022-2024, ICML 2023-2024, ICLR 2024-2025