Hoyong Kim

2021.03 ~ 2026.02 Ph.D of Artificial Intelligence, College of Information and Computing, Department of AI Convergence, Gwangju Institute of Science and Technology (GIST)
2018.03 ~ 2020.08 Master of Big-Data Science, Department of Big-Data Science, UST-KISTI Campus
2014.03 ~ 2018.02 Bachlor of Software, Department of Software, Ajou University

Research Interests

Neural Collapse

Analyze the model's behavior during/after training by focusing on the relation ship between the last-layer features and the class weight vectors, which are converged to a simplex ETF

Imbalanced Learning

Improve the model performance on minority classes while maintaining other classes' accuracies by mitigating the minority collapse

Continual Learning

Mitigate the catastrophic forgetting in continual learning by alleviating the class-wise interference in last-layer features and classifier

Research Experience

Research Assistant at IRR Lab in AI Graduate School, GIST

Analyze the search space of neural networks
- Find the effective search space based on the assumption that there is an effective search space when optimizing classification models and restricting the models to converge to it is useful to training

Sep. 2020 ~ Feb. 2021

Research Student at Research Data Sharing Center in Div. of National S&T Data, KISTI

Develop Clustering Module for Issue Analysis from News Articles
Improve the performance of Clustering Module in terms of speed and accuracy
- Make Gold Standard Dataset of Synonym in Korean from ‘Korea Open dictionary’
- Construct Deep Learning Model based on Siamese Neural Network for measuring Cosine Similarity between synonyms

Mar. 2019 ~ Aug. 2020

Research Intern at DILAB in Department of Software Convergence Engineering, Kunsan National University

Classification of Web Documents for people with the same name
- When searching specific person on search engine, classify web documents whether the document is about that person or not.
Find Trouble makers from E-mail Record Data
- Gather the E-mail data from Company DB, analyze how many times he or she sends and receives e-mails about each person and find some Trouble makers from that analysis.
Emotional analysis on News Articles about Hyun-dai Car.
- Gather news articles from web site by using web crawler, and analyze whether the article about Hyun-dai Car is positive or negative.
Deep Learning Model for Traffic Accident Points Prediction
- By using about 10,000 sensing data about the Traffic Accidents, which occurred in Daegu, Korea, predict the place where traffic accidents will be occurred by using Feedforward Neural Network.

Jul. 2015 ~ Feb. 2018

Publications

Conferences

[C12] Balancing Mixed Labels: Mixup meets Neural Collapse in Imbalanced Learning

Hoyong Kim, and Kangil Kim*

ICLR'2026, under-review OpenReview

CV / imbalanced learning / data augmentation / nerual collapse

[C11] Revisiting Softmax Masking: Stop Gradient for Enhancing Stability in Replay-based Continual Learning

Hoyong Kim, Minchan Kwon, and Kangil Kim*

The Third Conference on Lifelong Learning Agents (CoLLAs'2024 Workshop track), poster paper / video / arXiv

CV / continual learning

[C10] Asymptotic Midpoint Mixup for Margin Balancing and Moderate Broadening

Hoyong Kim, Semi Lee, and Kangil Kim*

The Third Conference on Lifelong Learning Agents (CoLLAs'2024 Workshop track), poster paper / video / arXiv

CV / data augmentation

[C9] Fixed Non-negative Orthogonal Classifier: Inducing Zero-mean Neural Collapse with Feature Dimension Separation

Hoyong Kim and Kangil Kim*

The Twelfth International Conference on Learning Representations (ICLR'2024), poster paper / video

CV / nerual collapse / imbalanced learning / continual learning

[C8] Spherization Layer: Representation Learning Using Only Angles

Hoyong Kim and Kangil Kim*

Advances in Neural Information of Processing Systems 35 (NeurIPS'2022), poster paper / video

CV / NLP / hyperspherical learning

[C7] Learning from matured dumb teacher for fine generalization

HeeSeung Jung, Kangil Kim, Hoyong Kim, and Jong-Hun Shin*

arXiv 2021, preprint arXiv

NLP / knowledge distillation

[C6] Dprelu: Dynamic parametric rectified linear unit

Kien Mai Ngoc, Donghun Yang, Iksoo Shin, Hoyong Kim, and Myunggwon Hwang*

The 9th International Conference on Smart Media and Applications (SMA'2020), poster paper

CV / activation function

[C5] Out-of-Distribution Detection Based on Distance Metric Learning

Donghun Yang, Iksoo Shin, Kien Mai Ngoc, Hoyong Kim, Chanhee Yu, and Myunggwon Hwang*

The 9th International Conference on Smart Media and Applications (SMA'2020), poster paper

CV / out-of-distribution / distance metric learning

[C4] Hierarchical Clustering of News Articles for Semi-Real-time Issue Analysis

Hoyong Kim, Hongjun Jang, Seungwoo Lee, and Dongmin Seo*

Korean Institute of Information Scientists and Engineers (KSC'2019), poster (in Korean) paper

NLP / application / data preprocessing / clustering / word embedding

[C3] Development of Web Crawler and Network Analysis Technology for Occurrence and Prediction of Flooding

Dongmin Seo, Hoyong Kim, Jeongha Lee, and Seokhwan Hwang*

Korea Contents Association (KCA'2019), poster (in Korean) paper

Tabular / application / data preprocessing

[C2] Prediction of Korean Basketball League playoff power-ranking based on Recurrent Neural Networks using KBL Efficiency

Seung-Kyo Jin, Hoyong Kim, and Sa-kwang Song*

Korea Contents Association (ICCC'2019), poster - Best Paper Award paper

Tabular / application / data preprocessing / RNN

[C1] A Word Semantic Similarity Measure Model using Korean Open Dictionary

Hoyong Kim, Min-Ho Lee, and Dongmin Seo*

Korea Contents Association (KSA'2018), oral - Best Paper Award (in Korean) paper

NLP / word embedding / data preprocessing

Journals

[J2] Hierarchical and Incremental Clustering for Semi-Real-time Issue Analysis on News Articles

Hoyong Kim, Seungwoo Lee, Hongjun Jang, and Dongmin Seo*

The Journal of Korea Contents Association, 2020 (in Korean) paper

NLP / application / data preprocessing / clustering / word embedding

[J1] Feature selection and visualization based on expectation maximization and principal component analysis for traffic accidents point prediction

Hoyong Kim, Jeong-Jae Kim, Sang-Min Park, In-Hwan Song, and Byung-Won On*

The Journal of the HCI Society of Korea, 2019 (in Korean) paper

Tabular / application / data preprocessing

Honors and Awards

RA Student Research Achievement Scholarship, AI Graduate School, GIST, 2023, 2024
Minister’s Commendation, The 3rd Government Innovation Proposal Hackaton, Ministry of the Interior and Safety, 2020
Excellence Prize, 2020 Campus Patent University, KIPA, 2020
Best Paper Award, International Conference on Convergence Contents, Korea Contents Association, 2019
Best Paper Award, KCA 2018 Spring Conference, Korea Contents Association, 2018
Grand Prize, Ajou Greative Software Festival, Ajou Univ., 2016

Academic Services

Reviewer Service: ICLR'2025, NeurIPS'2025, ICLR'2026

Page updated

Google Sites

Report abuse