Search

Taehyeon Kim

 DL Research Scientist
 Ph. D. Candidate @ KAIST AI [OSI Lab]
 Advised by Prof. Se-Young Yun
 Located in Seoul, South Korea
Education
 M.S. in Data Science (KAIST)
 B.S. in Mathematical Science (KAIST)
 Minor in Intellectual Property (KAIST)
My expected graduation date is Aug. 2024.  Contact: potter32@kaist.ac.kr, kimtaehyeon610@gmail.com (permanant)

 About

My goal is to tackle trustworthy and real-world AI/ML challenges. Specifically, my interests include:
•
Knowledge Distillation & Learning with Noisy Labels
•
Federated Learning & AutoML & Semi-Supervised Learning
•
Optimization for training deep neural networks (Efficient Deep Learning)
•
Language Model & Instruction Tuning
•
Game-Changing Research

️ News

Oct. 2023
May. 2023
Jan. 2023
 Working with Google Research NYC
 Invited to Liner Dev conference
 Working with DynamoFL Research Team
Dec. 2022
 4th Award in NeurIPS2022 Competition-Weather4Cast [Link]
 Two papers accepted to NeurIPS2022 Workshop
 Invited talk at HyperConnect Research Team
 Invited talk at KSC Conference
 Invited to Twelve Labs Networking
Oct. 2022
 Qualcomm Innovation Fellowship Winner in 2022 [Link]
Jul. 2022
 Two papers accepted to ICML2022 Workshop (including an oral presentation)
~ Jun. 2022
Previous News

Working at/with

Google Research

NYC, U.S.A.
Oct. 2023 - Dec. 2023

DynamoFL

(YC W22)
U.S.A. (Remote)
Jan. 2023 - May. 2023

Research Intern

•
Research on Language Models.

Research Intern

•
Research on real-world federated learning problems.
•
Semi-supervised federated learning

OpenMined

England (Remote)
Jun. 2022 - Jun. 2023

Research Scientist

•
Social-focused AI research: privacy, fairness, accountability, and transparency.

Qualcomm

Seoul, South Korea
Jun. 2021 - Dec. 2021

CV & ML Ph.D. Internship for Autonomous Driving report

•
Designing a resource-efficient and accurate backbone for ADAS
•
1 Paper and 1 US Patent
•
Knowledge Distillation, Neural Architecture Search

Data Mining Lab

Daejeon, South Korea
Mar. 2017 - Feb. 2018

Undergraduate Research Internship

•
Data Science, KAIST (Advisor: Jae-Gil Lee)
•
Trajectory prediction based on user card data

 Publications & Technical Reports

Search
Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Approach for Object Detection
NeurIPS2023
Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Approach for Object Detection
NeurIPS2023
Region-Conditioned Orthogonal 3D U-Net for Weather4Cast Competition
Competition
NeurIPS2022
Region-Conditioned Orthogonal 3D U-Net for Weather4Cast Competition
Competition
NeurIPS2022
Revisiting the Activation Function for Federated Image Classification
Workshop
NeurIPS2022
Revisiting the Activation Function for Federated Image Classification
Workshop
NeurIPS2022
Layover Intermediate Layer for Multi-Label Classification in Efficient Transfer Learning
Workshop
NeurIPS2022
Layover Intermediate Layer for Multi-Label Classification in Efficient Transfer Learning
Workshop
NeurIPS2022
Revisiting Architecture-aware Knowledge Distillation: Smaller Models and Faster Search
Workshop
ICML2022
Revisiting Architecture-aware Knowledge Distillation: Smaller Models and Faster Search
Workshop
ICML2022
Supernet Training for Federated Image Classification under System Heterogeneity
Workshop
ICML2022
Supernet Training for Federated Image Classification under System Heterogeneity
Workshop
ICML2022
Revisiting Kernel Orthogonality Regularization: A Study for Convolutional Neural Networks in Image Classification
IEEE Access
Revisiting Kernel Orthogonality Regularization: A Study for Convolutional Neural Networks in Image Classification
IEEE Access
FINE Samples for Learning with Noisy Labels
NeurIPS2021
FINE Samples for Learning with Noisy Labels
NeurIPS2021
Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillation
IJCAI2021
Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillation
IJCAI2021
Adaptive Local Bayesian Optimization Over Multiple Discrete Variables
Competition
NeurIPS2020
Adaptive Local Bayesian Optimization Over Multiple Discrete Variables
Competition
NeurIPS2020
Efficient Model for Image Classification With Regularization Tricks
PMLR
Competition
NeurIPS2019
Efficient Model for Image Classification With Regularization Tricks
PMLR
Competition
NeurIPS2019
Orthogonal feature regularization : a novel approach for training robust models, Korea Advanced Institute of Science and Technology
Master Thesis
KAIST
Orthogonal feature regularization : a novel approach for training robust models, Korea Advanced Institute of Science and Technology
Master Thesis
KAIST
Accurate and Fast Federated Learning via Combinatorial Multi-Armed Bandits.
Arxiv
Technical Report
Accurate and Fast Federated Learning via Combinatorial Multi-Armed Bandits.
Arxiv
Technical Report
Benchmark Dataset for Precipitation Forecasting by Post-Processing the Numerical Weather Prediction
Arxiv
Benchmark Dataset for Precipitation Forecasting by Post-Processing the Numerical Weather Prediction
Arxiv
Mold into a Graph: Efficient Bayesian Optimization over Mixed-Spaces
Arxiv
Mold into a Graph: Efficient Bayesian Optimization over Mixed-Spaces
Arxiv

Preprints (Under Review)

 Tackling Heterogeneous Label Noise in Federated Learning with Label-Mixture Regularization
Taehyeon Kim*, Donggyu Kim*, Se-Young Yun
 Overcoming Heterogeneity in Federated Learning with Supernet Training and Parametric Normalization
Taehyeon Kim, Se-Young Yun
 A Survey of Supernet Optimization and its Applications: Spatial and Temporal Optimization for Neural Architecture Search
Stephen Cha, Taeheyeon Kim, Hayeon Lee, Se-Young Yun
 Task Adaptive Distillation in Cross Domain Few Shot Learning
Yongjin Yang, Taehyeon Kim, Se-Young Yun

 Details

Students
Skills
Leadership
Awards & Achievements
Research Project Experience
Scholarship & Fellowship
Publications Details