Jinyoung Kim

Jinyoung Kim

CSE PhD student @ UMich

About Me

Hello! I am a first-year Ph.D. student in Computer Science and Engineering at the University of Michigan, where I’m advised by Prof. Lu Wang. Previously, I was an undergraduate researcher at the Vision & Learning Lab at Seoul National University, advised by Prof. Gunhee Kim.

My primary research interests lie in natural language processing and machine learning, with a focus on developing adaptive and trustworthy AI systems that can operate reliably in dynamic environments. I work on:

  • Learning-to-critique with reinforcement learning for improving non-verifiable and long-horizon reasoning
  • Retrieval-augmented and data-centric adaptation for models facing evolving knowledge
  • Fine-grained evaluation frameworks that enable continual improvement and robust generalization

Here is my CV.

Current Research Projects

1. Learning-to-Critique with Reinforcement Learning

I explore reinforcement learning methods that enable models to critique, revise, and refine their own reasoning, especially in non-verifiable tasks where correctness is difficult to judge. My work investigates how models can generate structured feedback, self-evaluate intermediate steps, and improve through iterative critique loops, with the goal of building more reliable and interpretable reasoning systems.

2. Retrieval-Augmented and Data-Centric Adaptation

I study how AI systems can incorporate new or evolving knowledge through retrieval and data-centric methods. This includes understanding when models should rely on external information, how to handle shifting or conflicting sources, and how retrieval affects downstream reasoning. My research aims to improve robustness as real-world knowledge changes, without requiring expensive retraining.

3. Fine-Grained Evaluation for Continual Improvement

I develop evaluation frameworks that diagnose model behavior at a detailed level, capturing nuances in reasoning quality, factuality, and robustness across domains. These frameworks support continual model improvement by revealing where models struggle, what types of errors they make, and how their performance evolves under distribution shift.

Publications

  • When Should Dense Retrievers Be Updated in Evolving Corpora? Detecting Out-of-Distribution Corpora Using GradNormIR

    Dayoon Ko, Jinyoung Kim, Sohyeon Kim, Jinhyuk Kim, Jaehoon Lee, Seonghak Song, Minyoung Lee, Gunhee Kim
    ACL 2025 Findings
    [paper] [code]

  • DynamicER: Resolving Emerging Mentions to Dynamic Entities for RAG

    Jinyoung Kim, Dayoon Ko, Gunhee Kim
    EMNLP 2024
    [paper] [code]

  • GrowOVER: How Can LLMs Adapt to Growing Real-World Knowledge?

    Dayoon Ko, Jinyoung Kim, Hahyeon Choi, Gunhee Kim
    ACL 2024
    [paper] [code]

Education

  • University of Michigan (Aug. 2025 - Present)

    Ph.D. student in Computer Science and Engineering

  • Seoul National University (Mar. 2018 - Aug. 2024)

    Leave of absence for military service: Jul. 2020 - Jan. 2022
    B.S. in Computer Science and Engineering & B.A. in Economics
    Graduated with Summa Cum Laude

Experiences

  • Vision & Learning Lab, SNU (Sep. 2023 - Present)

    Research Intern

    • Conducted research on evolving knowledge and retrieval-augmented generation
    • Published 3 papers in top-tier conferences

  • CLOVA Voice & Avatar Team, NAVER (Jan. 2023 - Feb. 2023)

    Machine Learning Researcher

    • Topic: Unified Accent Estimation for Speech Synthesis
    • Achieved 7.8% improvement in Accent Phrase (AP) prediction accuracy by unifying estimation for AP and Accent Nucleus (AN) boundaries
    • Designed and implemented novel AN decoder framework to address challenges in the long-tail distribution of accent estimation

  • Music & Audio Research Group, SNU (Jul. 2022 - Aug. 2022)

    Research Intern

    • Conducted research on personalized symbolic music generation
    • Developed contrastive learning-based encoder and holistic frameworks for emotion and style conditioning in music generation

Leadership Experiences