I am a Ph.D. student in Electrical Engineering at KAIST, advised by Prof. Changick Kim.

My research interests include Test-time optimization and Neural Memory. . I particularly enjoy short-term performance tuning challenges , where rapid iteration and optimization are critical.

I am a multi-time challenge winner , including:
๐Ÿฅ‡ CVPR 2022 BMTT Challenge โ€“ Winner
๐Ÿฅ‡ NeurIPS VisDA 2022 Challenge โ€“ Winner
๐Ÿฅ‡ NeurIPS Weather4Cast 2022 Challenge โ€“ Winner
๐Ÿฅ‡ ICCV GeoNet 2023 Challenge โ€“ Winner

Currently, I am not collaborating on any topics other than weather forecasting and test-time optimization.

Contact

  • msseok96 [at] gmail.com

    minseok.seo [at] kaist.ac.kr

  • 291 Daehak-ro, Yuseong-gu, Daejeon 34141, Republic of Korea

Education

  • PhD in EE, KAIST, 2025~

    on "Vision Foundation Model (TODO)"

  • MS in EE, KAIST, 2023

    on "Domain Adaptation"

  • MS in EE, Hanbat National University, 2021

    on "Active Learning for AI application"

  • BS in EE, Hanbat National University, 2020

Academic Activities

  • Reviewer prize ICCV 2023, ICML 2026
  • AC in ICLR, 2025, CVPR2025, ECCV2026
  • Program Committee in AAAI2025
  • Reviewer at CVPR, NeurIPS, ICLR, ICCV, ECCV, ICML, AAAI, TNNLS, TIP
  • Invited talk at the Industry Technology Workshop ICME 2026, AI friends

Research Experiences

  • NVIDIA (INTERN) , London, UK
    July 2026 - Sep 2026

    Research Scientist, NVIDIA
  • SI-Analytics (military service) , Daejeon, Korea
    Mar 2022 - Sep 2025

    Research Scientist, SI-Analytics

An open research note

Research should invite conversation, not just accumulate citations.

AI papers are appearing faster than ever, and AI-generated reviews are becoming part of the process. I believe the most useful response is not simply more volume, but more visible, thoughtful human dialogue.

For my new work, I plan to share an archival version openly, invite a small number of researchers whose judgment I deeply respect to read it, and publish their perspectives alongside the paperโ€”with their permission, in their own words, and without editing away disagreement.

01 Archive openly

Make the work, evidence, and limitations accessible to everyone.

02 Invite trusted readers

Ask people whose taste and judgment I respect for candid feedback.

03 Publish the dialogue

Share their perspectives so readers can evaluate the work with richer context.

Reviews alongside the work

Trusted Perspectives

Selected archival papers below include invited perspectives from researchers I admire. These notes may support, question, or challenge the work; that independence is the point.

These are invited personal perspectives, published with permission. They are not anonymous or institutional peer review.

Publications

    Vision Foundation Model

  • Efficient Test-Time Optimization for Depth Completion via Low-Rank Decoder Adaptation

    Minseok Seo*, Wonjun Lee*, Jaehyuk Jang, Changick Kim

    arXiv 2026

    [ paper / code ]

    Trusted Perspectives ยท invitations in progress

    Invited reviews will appear here with each reviewer's name, affiliation, date, and permission.

  • Visual Self-Supervised Learning in Practice: A Survey of Alignment-, Reconstruction-, and Prediction-Based Paradigms

    Minseok Seo, Inyong Koo, Young-tack Oh, Young-Jae Park, Dong-geol Choi, Hae-Gon Jeon, Changick Kim

    arXiv 2026

    [ paper / code ]

    Trusted Perspectives ยท invitations in progress

    Invited reviews will appear here with each reviewer's name, affiliation, date, and permission.

  • Upsample Anything: A Simple and Hard to Beat Baseline for Feature Upsampling

    Minseok Seo, Mark Hamilton, Changick Kim

    CVPR 2026 (Compute Gold Star / Transparency Champion)

    [ paper / code ]

  • VideoTitans: Scalable Video Prediction with Integrated Short-and Long-term Memory

    Young-Jae Park*, Minseok Seo*, Hae-Gon Jeon

    NeurIPS 2025

    [ paper / code ]

    • Weather AI

  • RainODE: Continuous-Time Precipitation Forecasting with Latent Neural ODEs

    Yeeun Seong*, Doyi Kim*, Minseok Seo, and Changick Kim

    ECCV 2026

    [ paper / code ]

  • Station2Radar: queryโ€‘conditioned gaussian splatting for precipitation field

    Doyi Kim, Minseok Seo, Changick Kim

    ICLR 2026

    [ paper / code ]

  • Data-driven Precipitation Nowcasting Using Satellite Imagery

    Young-Jae Park, Doyi Kim, Minseok Seo, Hae-Gon Jeon, Yeji Choi

    AAAI 2025 (ORAL)

    [ paper / code ]

  • Probabilistic Weather Forecasting with Deterministic Guidance-based Diffusion Model

    Donggeun Yoon* , Minseok Seo , Doyi Kim, Yeji Choi, and Donghyeon Cho

    ECCV 2024

    [ paper / code ]

  • Long-Term Typhoon Trajectory Prediction: A Physics-Conditioned Approach Without Reanalysis Data

    Young-Jae Park*, Minseok Seo*, Doyi Kim, Hyeri Kim, Sanghoon Choi, Beomkyu Choi, Jeongwon Ryu, Sohee Son, Hae-Gon Jeon, Yeji Choi

    ICLR 2024 (Spotlight)

    [ paper / code ]

  • Masked Autoregressive Model for Weather Forecasting

    Doyi Kim*, Minseok Seo*, Hakjin Lee, Junghoon Seo

    arXiv 2024

    [ paper / code ]

  • Weather4cast at neurips 2022: Super-resolution rain movie prediction under spatio-temporal shifts

    Weather4Forecast (Winner)

    NeurIPS 2022 competition track

    [ paper / code ]

    • Application

  • Self-Pair: Synthesizing Changes From Single Source for Object Change Detection in Remote Sensing Imagery

    Minseok Seo, Hakjin Lee, Yongjin Jeon, Junghoon Seo

    WACV 2023

    [ paper / code ]

  • Bidirectional domain mixup for domain adaptive semantic segmentation

    Daehan Kim*, Minseok Seo*, Kwanyong Park, Inkyu Shin, Sanghyun Woo, In So Kweon, Dong-Geol Choi

    AAAI 2023 (ORAL)

    [ paper / code ]

  • VisDA 2022 challenge: Domain adaptation for industrial waste sorting

    VisDA (Winner)

    NeurIPS 2022 Competition Track

    [ paper / code ]

  • Unsupervised change detection based on image reconstruction loss

    Hyeoncheol Noh*, Jingi Ju*, Minseok Seo*, Jongchan Park, Dong-Geol Choi

    CVPR 2022

    [ paper / code ]

  • Pt4al: Using self-supervised pretext tasks for active learning

    John Seon Keun Yi*, Minseok Seo*, Jongchan Park, Dong-Geol Choi

    ECCV 2022

    [ paper / code ]

  • A self-supervised sampler for efficient action recognition: Real-world applications in surveillance systems

    Minseok Seo, Donghyeon Cho, Sangwoo Lee, Jongchan Park, Daehan Kim, Jaemin Lee, Jingi Ju, Hyeoncheol Noh, Dong-Geol Choi

    RAL-ICLR 2021

    [ paper / code ]

  • OCR-based Inventory Management Algorithms Robust to Damaged Images

    Minseok Seo, Daehan Kim, Hyeyoon Kang, Donghyeon Cho, Dong-Geol Choi

    ICRA 2021

    [ paper / code ]

Research Collaboration