Jihoon Tack

Contact: jihoontack@{gmail.com,microsoft.com}

About

I am a Senior Researcher in the AI Interaction and Learning group at Microsoft Research Redmond. My research focuses on building AI agents that can collaborate with humans to solve complex, long-horizon tasks. In particular, I am interested in developing scalable RL frameworks and environments to train agents to understand user intent, plan over long horizons, and collaborate through interaction.

Before joining Microsoft Research, I completed my Ph.D. at KAIST under the supervision of Jinwoo Shin, where I worked on meta-learning and efficient adaptation of large-scale models. I was a recipient of Google Ph.D. Fellowship in machine learning.

Publications

* denotes equal contribution

LLMs Get Lost in Evolving User Intent

Persona-Pruner: Sculpting Lightweight Models for Role-Playing

Beyond Correctness: Learning Robust Reasoning via Transfer

Scalable and Robust LLM Unlearning by Correcting Responses with Retrieved Exclusions

LLM Pretraining with Continuous Concepts

Think Clearly: Improving Reasoning via Redundant Token Pruning

Mamba Drafters for Speculative Decoding

ReGUIDE: Data Efficient GUI Grounding via Spatial Reasoning and Search

ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification

Learning to Contextualize Web Pages for Enhanced Decision Making by LLM Agents

Sparsified State-Space Models are Efficient Highway Networks

Online Adaptation of Language Models with a Memory of Amortized Contexts

ReMoDetect: Reward Models Recognize Aligned LLM's Generations

Optimized Feature Generation for Tabular Data via LLMs with Decision Tree Reasoning

Tabular Transfer Learning via Prompting LLMs

Unleashing the Power of Meta-tuning for Few-shot Generalization Through Sparse Interpolated Experts

Learning Large-scale Neural Fields via Context Pruned Meta-Learning

Modality-Agnostic Self-Supervised Learning with Meta-Learned Masked Auto-Encoder

Modality-Agnostic Variational Compression of Implicit Neural Representations

STUNT: Few-shot Tabular Learning with Self-generated Tasks from Unlabeled Tables

Rethinking the Entropy of Instance in Adversarial Training

Meta-Learning with Self-Improving Momentum Target

K-centered Patch Sampling for Efficient Video Recognition

Generating Videos with Dynamics-aware Implicit Generative Adversarial Networks

Consistency Regularization for Adversarial Robustness

Meta-Learning Sparse Implicit Neural Representations

CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances

Adversarial Self-Supervised Contrastive Learning

Experience

Senior Researcher, Microsoft Research, AI Interaction and Learning Group, Nov. 2025 - Present
Research Scientist Intern, Meta FAIR (Host: Xian Li), Jun. 2024 - Mar. 2025
Visiting Student / Collaborator, University of Oxford (Host: Yee Whye Teh), Jun. 2023 - May 2024
External Collaborator, Jonathan Richard Schwarz at DeepMind → Thomson Reuters, Jun. 2022 - May 2024
External Collaborator, Jaeho Lee at POSTECH, Apr. 2022 - May 2023

Education

Ph.D. in Artificial Intelligence
Korea Advanced Institute of Science and Technology (KAIST), Mar. 2019 - Aug. 2025
Advisor: Jinwoo Shin

B.S. in Electrical Engineering and Mathematics (minored)
Korea Advanced Institute of Science and Technology (KAIST), Mar. 2014 - Feb. 2019

Honors and Awards

Qualcomm Innovation Fellowship Korea 2024
Google Ph.D. Fellowship 2023, Research Area: Machine Learning
Samsung Humantech Paper Awards 2023, Bronze Prize
Best Paper Award, Korean Artificial Intelligence Association 2021
Qualcomm Innovation Fellowship Korea 2020
Conference Scholar Award: NeurIPS 2023
Reviewer Awards: NeurIPS 2020, CVPR 2021, NeurIPS 2022, ICLR 2025

Academic Activities

Conference Reviewer: NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, AAAI, ACL, EMNLP
Journal Reviewer: TMLR, IJCV, IEEE TPAMI, IEEE TNNLS, IEEE TIFS, IEEE TIP