Search

Taehyeon Kim

 Prev: Google Research (NYC), Qualcomm AI, DynamoFL (YCW22)
 PhD Student @ KAIST AI [OSI Lab]
Education
 PhD - KAIST AI
 MS - KAIST Data Science
 BS - KAIST Mathematical Science
[Minor-Intellectual Property]
I am a Ph.D. Student in Optimization & Statistical Inference (OSI) Lab @ KAIST, advised by Prof. Se-Young Yun. I worked as a PhD Intern @ Google Research, Qualcomm AI, and DynamoFL (YCW22). My expected graduation date is Feb. 2025.  Contact: potter32 [at] kaist.ac.kr, kimtaehyeon610 [at] gmail.com (permanent)
I am in the job market! Please feel free to contact me!  Click CV & LinkedIn (Updated: Oct, 2024)!

 About

My goal is to tackle trustworthy and real-world AI/ML challenges
•
LLM Inference: Instructive Decoding, Speculative Decoding, Parallel Decoding
•
Alignment Algorithm: Knowledge Distillation & Learning with Noisy Labels
•
Data Heterogeneity: Federated Learning & AutoML & Semi-Supervised Learning
•
Game-Changing Research

 CV (Updated: Oct 10, 2024)

️ News

Oct. 2024
 1 Accepted @ NeurIPS2024W: Speculative Decoding with multiple drafters
Sep. 2024
 Successfully passed my PhD Proposal!
 1 Accepted @ EMNLP2024 Main - Specialized Speculative Decoding!
 2 Accepted @ NeurIPS 2024 - Speculative Decoding and Block Transformer!
Jul. 2024
🤝🏻
PR - LLM Inference @ Efficiency Seminar (✅ Click to see 👈🏻)
Jun. 2024
 1 Accepted @ ICML2024W: Blockwise Parallel Decoding (Speculative Decoding)
🤝🏻
PR - Federated Object Detection @ KCC Seminar (✅ Click to see 👈🏻)
May. 2024
 Attending ICLR 2024 @ Vienna, Aus  
Jan. 2024
 1 Accepted @ ICLR2024 (Spotlight): Instruction Following on Large Language Model
Dec. 2023
 1 Accepted @ NeurIPS2023W: Instruction Tuning & Instruction Following
 1 Accepted @ NeurIPS2023: Semi-Supervised Federated Object Detection
 Attending NeurIPS 2023 @ NOLA, US  
 1 Accepted @ AAAI2024: Few-shot & Domain Generalization
Oct. 2023
 Working with Google Research NYC

Working at/with

Search
Name
When
Research
Advisor/Coworker
Google Research (NYC, U.S.A. ) - PhD Intern
Open
2023.10 - 2023.12
Parallel Decoding
Language Model
Adrian Benton
Ananda Theertha Suresh
Kishore Papineni
DynamoFL (SF, U.S.A.) - PhD Intern
Open
2023.01 - 2023.05
Semi-Supervised Object Detection
Federated Learning
Eric Lin
Qualcomm (Seoul, Korea) - PhD Research Intern
Open
2021.06 - 2021.12
Neural Architecture Search
Knowledge Distillation
Heesoo Myeong
Data Mining Lab, KAIST (Daejeon, Korea) - Undergrad Intern
Open
2017.03 - 2018.02
Trajectory Prediction
Jaegil Lee

 Publications & Technical Reports

Search
Title
Conference and Journal
Author
A Unified Framework for Speculative Decoding with Multiple Drafters as a Bandit
Open
NeurIPS2024
Workshop
1st author
Exploring and Improving Drafts in Blockwise Parallel Decoding
Open
NeurIPS2024
1st author
Block Transformer: Global-to-Local Language Modeling for Fast Inference
Open
NeurIPS2024
Co-author
Towards Fast Multilingual LLM Inference: Speculative Decoding and Specialized Drafters
Open
EMNLP2024
1st author
Instructive Decoding: Instruction-Tuned Large Language Models are Self-Refiner from Noisy Instructions
Open
ICLR2024
Spotlight
NeurIPS2023
Workshop
1st author
Leveraging Normalization Layer in Adapters With Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning
Open
AAAI2024
Co-author
Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Federated Object Detection
Open
NeurIPS2023
1st author
Weather4cast at NeurIPS 2022: Super-Resolution Rain Movie Prediction under Spatio-temporal Shifts
Open
NeurIPS2022
Competition
Co-author
Region-Conditioned Orthogonal 3D U-Net for Weather4Cast Competition
Open
NeurIPS2022
Competition
1st author
Revisiting the Activation Function for Federated Image Classification
Open
NeurIPS2022
Workshop
Co-author
Layover Intermediate Layer for Multi-Label Classification in Efficient Transfer Learning
Open
NeurIPS2022
Workshop
Co-author
Revisiting Architecture-aware Knowledge Distillation: Smaller Models and Faster Search
Open
ICML2022
Workshop
1st author
Supernet Training for Federated Image Classification under System Heterogeneity
Open
ICML2022
Workshop
1st author
Revisiting Kernel Orthogonality Regularization: A Study for Convolutional Neural Networks in Image Classification
Open
IEEE Access
1st author
FINE Samples for Learning with Noisy Labels
Open
NeurIPS2021
1st author
Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillation
Open
IJCAI2021
1st author
Adaptive Local Bayesian Optimization Over Multiple Discrete Variables
Open
NeurIPS2020
Competition
1st author
Efficient Model for Image Classification With Regularization Tricks
Open
NeurIPS2019
Competition
1st author
Orthogonal feature regularization : a novel approach for training robust models, Korea Advanced Institute of Science and Technology
Open
Master Thesis
KAIST
1st author
Accurate and Fast Federated Learning via Combinatorial Multi-Armed Bandits.
Open
Arxiv
Technical Report
1st author
Benchmark Dataset for Precipitation Forecasting by Post-Processing the Numerical Weather Prediction
Open
Arxiv
Technical Report
1st author
Mold into a Graph: Efficient Bayesian Optimization over Mixed-Spaces
Open
Arxiv
Co-author
A Survey of Supernet Optimization and its Applications: Spatial and Temporal Optimization for Neural Architecture Search
Open
Arxiv
Technical Report
Co-author
COUNT23

Preprints (Under Review)

 Hypernetwork-Driven Model Fusion for Federated Domain Generalization
Marc Bartholet*, Taehyeon Kim*, Ami Beuret, Se-Young Yun, Joachim M. Buhmann
 FLR: Label-Mixture Regularization for Federated Learning with Noisy Labels
Taehyeon Kim, Donggyu Kim, Se-Young Yun

 Leadership Awards  Activities

Search
Title
When
Role
General Manager @ NIMS Project (Rainfall Forecast; 5 Labs)
Open
2022
Leadership
Best Presentation Award @ KCC Conference
Open
2022
Award
5th Award @ NeurIPS 2022 Weather4cast
Open
2022
Award
Winner @ Qualcomm AI Fellowship
Open
2022
Award
Best Poster Awards @ KAIST AI 21/22 Workshop
Open
2022
Award
AI Lecturer @ Curinc (Undergrad UCB, Columbia, Boston)
Open
2021
Lecturer
Lecturer @ LG CNS, LG Electronics
Open
2021
2020
2019
Lecturer
Representative of doctoral students @ KAIST AI
Open
2021
Leadership
Advisory Committee @ National Science Museum
Open
2020
Advisor
8th Award @ NeurIPS 2020 Black-Box Optimization Challenge
Open
2020
Award
Student Leader @ OSI Lab
Open
2020
Leadership
2nd, 3rd Awards @ NeurIPS 2019 MicroNet Challenge
Open
2019
Award
Lecturer @ Samsung SW Academy
Open
2019
Lecturer
Lecturer @ KAIST Leadership - Regular Course
Open
2017
Lecturer
1st Award @ Universiade
Open
2015
Award
Vice Captain of Cheerleading @ KAIST
Open
2015
Leadership
COUNT16

 Research Projects

Search

 Invited Talks

Search

 Services & Others

Search

Top Conference

NeurIPS {2022, 2023, 2024}, ICML {2022, 2023, 2024}, ICLR {2024}, ECCV 2022, KDD 2024, ACL 2024

Journal

Pattern Recognition
Academic Reviewer

Title: Surplus human @ AI with Weird Wonderland

•
22.03.10 - 22.03.16, CELINE PARK GALLERY
•
Organized by Next interface lab
•
Vania Oh, Junbeom Shin, Taehyeon Kim
AI in Art

 Others

Students (Mentee)
Publications Details
🫥
Blog Posts (Kor)
Research Posts (Eng)