Keumgang Cha
Keumgang Cha
My Keyword : Representation Learning, Multi Modality, Foundation Model, Large Scale Training, Machine Learning Infra Structure
Hello, I am conducting machine learning engineering and research at SI Analytics, a satellite / aerial image analysis company in Korea, and I am a team leader. I received my bachelor's and master's degrees at the Department of Mechanical Engineering at Pusan National University, majoring in robotics control. Current interest in the field of machine learning engineers is how to train very large models, and how to parallelize pipelines and models to make training and evaluation easier. In the field of machine learning research, I am interested in Representation Learning, Multi Modality, Foundation Model using various data. If you would like to contact me regarding this, please contact me via email or LinkedIn.
My Link
Google Scholar - https://scholar.google.com/citations?user=tLYtx20AAAAJ&hl=ko
LinkedIn - https://www.linkedin.com/in/keumgang-cha-58100b227/
github - https://github.com/chagmgang
portfolio - https://chagmgang.notion.site/Hi-Kevin-Cha-16f27bd251e7803eb6f8e05f5d61cb0a?pvs=4
email - chagmgang@gmail.com
Interested in
Machine Learning Engineering
Data/Model/Pipeline/Tensor Parallelism
Training / Inference acceleration in Deep Learning
Representation Learning
Representation Learning in Multi Modality, such as Text-Image and Hyperspectral Image.
Career
2021.07~Current, Machine Learning Engineer, Team Leader @ "SI Analytics".
2020.01~2021.06, Machine Learning Engineer and Researcher @ "Setrec Initative"
2018.01~2019.12, Machine Learning Researcher @ "PlanI"
2018.06~2021.06, Technical Reserach Personnel for Military Service
2016.07~2017.03, Robotics Researcher @ "KITECH"
2016.10~2017.09, Researcher @ "KAIST"
Education
2018.09~2019.12, Ph.D Course @ "KAIST"
2014.03~2016.02, Master Course @ "Pusan National University", (4.40 / 4.5)
2010.03~2014.02, B.S Course @ "Pusan National University", (3.80 / 4.5)
Side Project
DINO v2 Reimplementation for Remote Sensing with Vision Transformer
Reinforcement Learning with Distributed Architecture
PPO Reinforcement Learning Implementation
Project
2024.09 ~ 2025.01 Building Self Supervised Learning for Multi Sensor (EO, SAR) in Remote Sensing
Building Convolution Vision Transformer Model for Remote Sensing Multi Spectral Imagery(EO, SAR)
keywords : Hyper Spectral, Self Supervised Learning, Remote Sensing
Multi Node Multi GPU Training, Object Detection
2024.09 ~ 2024. 11, Buliding Pixel Level Computer Vision Foundation Model for Remote Sensing with transformers and Deepspeed.
Building Language Aware Foundation Panoptic Segmentation Model with transformers and Deepspeed.
keywords : Vision-Language Model, Panoptic Segmentation, Remote Sensing
2024.01 ~ 2024. 09, Buliding Foundation Object Detection Model for Remote Sensing with transformers and Deepspeed.
Building Language Aware Object Detection Model with transformers and Deepspeed.
keywords : Vision-Language Model, Object Detection, Remote Sensing
2023.04 ~ 2023.07, Building Remote Sensing Vision Language Contrastive Learning Model(CLIP) over 4B parameters
Building Vision Language CLIP Model in Remote Sensing with GPT2-xlarge and Vision Transformer 2.4B
keywords : Vision-Language Model, Representation Learning, Contrastive Learning
2023.01 ~ 2023. 07, Building large-scale model training/inference systems in computer vision areas such as object detection and semantic segmentation
Building Pipeline Parallelism to Rotated Object Detection and Semantic Segmention based on Vision Transformer in mmcv.
keywords : Machine Learning Engineering, Object Detection, Semantic Segmentation
2022.07 ~ 2022.12, Remote sensing field foundation model learning using large-scale unlabeled images
Building Billion-Scale Foundation Model in Remote Sensing Domain.
Building Large-Scale Unlabeled Remote Sensing Imagery.
keywords : Object Detection, Semantic Segmentation, Computer Vision Pretraining
2021.01~2022.07, Building MLOps System with MLFlow
2020.01~2021.06, Semantic Segmentation model and Rotated Object Detection model improvement using data analysis and collection
By building a data set with a wide range, IoU and F1 Score are dramatically improved in Semantic Segmentation and Rotated Object Detection, respectively.
2020.01~2021.06, Computer Vision Model Explainability Using Attention Algorithm
Using Self Attention Visualization in Computer Vision.
2018.01~2019.12, Content recommendation system using large-scale language processing system
Analyzing and formalizing the relationship between users or contents using various large-scale pre-learning language models and clustering methodologies.
Award
2nd, HAAFOR Challenge 2020, Natural Language Processing.
Article
Airflow tutorial blog - https://blog.si-analytics.ai/59?category=894442
Inference with pytorch and TensorRT, 1 - https://blog.si-analytics.ai/33?category=894442
Inference with pytorch and TensorRT, 2 - https://blog.si-analytics.ai/32?category=894442
Multi GPU Training with Horovod - https://blog.si-analytics.ai/22?category=894442
Multi GPU Training with Pytorch DataDistributedParallel - https://blog.si-analytics.ai/12?category=894442
Publication
On Pitfalls of Remove-And-Retrain: Data Processing Inequality Perspective, J Song, K Cha, J Seo, ArXiv, abs/2304.13836
Pushing the Limits of Vision-Language Models in Remote Sensing without Human Annotations, K Cha, D Yu, J Seo, arXiv preprint arXiv:2409.07048
A billion-scale foundation model for remote sensing images, K Cha, J Seo, T Lee, arXiv preprint arXiv:2304.05215
Contrastive multiview coding with electro-optics for sar semantic segmentation, K Cha, J Seo, Y Choi, IEEE Geoscience and Remote Sensing Letters 19, 1-5
RAIN-F+: The Data-Driven Precipitation Prediction Model for Integrated Weather Observations, Y Choi, K Cha, M Back, H Choi, T Jeon, Remote Sensing 13 (18), 3627
RAIN-F: A fusion dataset for rainfall prediction using convolutional neural network, Y Choi, K Cha, M Back, H Choi, T Jeon, 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS
Performance of Recommender Systems: Based on Content Navigator and Collaborative Filtering, KG Cha, SR Lee, JW Lee, SB Baik, arXiv preprint arXiv:1909.08219
A Study on Deep Learning Based Sauvegrain Method for Measurement of Puberty Bone Age, SB Baik, KG Cha, arXiv preprint arXiv:1809.06965
Movement Classification based on Acceleration Spectrogram with Dynamic Time Warping Method, B Noh, KG Cha, S Chang, 2017 18th IEEE International Conference on Mobile Data Management (MDM), 397-400
SPO based reaction force estimation and force reflection bilateral control of cylinder for tele-dismantling, KG Cha, SM Yoon, MC Lee, The Journal of Korea Robotics Society 12 (1), 1-10
Evaluation of a possibility of estimation of reaction force of simple hydraulic system using sliding perturbation observer, HC Lee, KG Cha, MC Lee, 2015 15th International Conference on Control, Automation and Systems
Estimated force based velocity synchronization for fly touch control in hot rolling process, SM Yoon, MC Lee, SJ Kim, HH Kim, KG Cha, 2015 IEEE International Conference on Advanced Intelligent Mechatronics
Reaction force estimation of hydraulic servo system using sliding perturbation observer, KG Cha, SM Yoon, HH Kim, KY Gim, MC Lee, 2015 IEEE International Conference on Advanced Intelligent Mechatronics
SMCSPO based Force Estimation for Jetting Rate Control of 3D Printer Nozzle to build a house, KG Cha, MC Lee, HJ Kim, Intelligent Robotics and Applications: 8th International Conference