About me
In the rapidly evolving world of AI, I am deeply interested in the next challenge we face: "How can we deploy AI safely?" To address this, I am dedicated to researching reliable and safe AI technologies. Specifically, my research focuses on uncertainty calibration and out-of-distribution detection.
[2024.07] One paper accepted in ECCV 2024
[2023.10] One paper accepted in NeurIPS 2023 Workshop on Distshift
[2023.09] Joined as a visiting scholar Carnegie Mellon University IITP AI Intensive Program (Aug. 2023 ~ Feb. 2024)
[2022.09] Joined CVML @CAU
Education
M.S. Department of Artificial Intelligence, CAU, 2024 (expected)
Advisor : prof. Junseok Kwon
B.S. Department of Computer Science, JBNU, 2022
Advisor : prof. Hyungki Kim
Research Interest
Safety & trustworthy AI
Uncertainty calibration
Out-of-distribution detection
Differential privacy
Publication (International)
Mijoo Kim, Junseok Kwon, “Uncertainty Calibration With Energy Based Instance-wise Scaling In the Wild Dataset”, ECCV, (2024)
Changdae Oh, Mijoo Kim, Hyesu Lim, Junhyeok Park, Euiseog Jeong, Zhi-Qi Cheng, Kyungwoo Song, “Towards Calibrated Robust Fine-Tuning of Vision-Language Models”, NeurIPS Workshop on Distribution Shifts, (2023) [pdf]
Hyungki Kim, Wonyong Lee, Mijoo Kim, Yoochan Moon, Taekyong Lee, Mincheol Cho, and Duhwan Mun, "Deep learning-based recognition of symbols and texts at an industrially applicable level from high-density piping and instrumentation diagram images", ESWA, (2021) [pdf]
Publication (Domestic)
Mijoo Kim, Jinhee Park, Younghyeon Park, Junseok Kwon, “Exploring Membership Inference Attacks in Diffusion Models”, Summer Annual Conference of IEIE, (2024)
Mijoo Kim, Junseok Kwon, “A Survey of Post-hoc Uncertainty Calibration Methods”, Summer Annual Conference of IEIE, (2023)
Wonyong Lee, Mijoo Kim, Duhwan Mun, Hyungi Kim, "Image format P&ID recognition technique using synthetic data and text-symbol Integrated detection", Korean Journal of Computational Design and Engineering, (2021)
Preprint
Changdae Oh*, Hyesu Lim*, Mijoo Kim, Dongyoon Han, Sangdoo Yun, Jaegul Choo, Alexander Hauptmann, Zhi-Qi Cheng, Kyungwoo Song, "Towards Calibrated Robust Fine-Tuning of Vision-Language Models", (2024) [pdf] under review
Experience
[2023.09 ~ 2024.02] Visiting scholar with IITP AI Intensive Program, Carnegie Mellon University
[2021.09 ~ 2021. 12] Vision engineer Internship at prior development team, NextLab
[2019.04 ~ 2021.08] Undergraduate Research Student at Visual Computing Lab, JBNU
Awards
[2022] Top prize at Korean Society for Computational Design and Engineering
[2021] Excellence Prize at Korean Society for Computational Design and Engineering Summer Conference
[2020] Excellence Prize at School Idea Hackathon
Patent
Mijoo Kim, Junseok Kwon. Uncertainty calibration system and method energy-based instance-specific scaling. KOR Patent 10-2024-0062366. filed May 13, (2024)
Chang Geun Lee, Jeong Gi Park, Mijoo Kim, Hyoung Seok Kim, Hak Jin Kim, Min Seo Park, and Seyoung Park. AI-Based Road Control System and Method Using Drones. KOR Patent 10-2022-0143599. filed November 1, (2022)
Projects more
JUMO (Development of a Beverage Serving Robot) @CMU
Robotics Jetson orin nano Jetbot serving robot SLAM speech recognition
Team Project in AI on Edge with Robotics
Oct. 2023 ~ Feb. 2024
Assembly of Jetbot (an open-source robot based on NVIDIA Jetson Nano)
Real-time slam navigation & mapping
Build speech recognition pipeline
Robot motion control
Towards Calibrated Robust Fine-Tuning of Vision-Language Models @CMU
Team research project in Large-Scale Multimedia Analysis 23f
Sep. 2023 ~ Dec. 2023
Vision-Language-model CLIP Uncertainty calibration Robust fine-tuning
reviewing research overview & literature search
draft writing & code writing & experiment
publication : [pdf]
Implementation of Diffusion Attentive Attribution Maps @CMU
Implementation project in Natural Language Processing 23f
Sep. 2023 ~ Dec. 2023
NLP Explainable AI Stable Diffusion Model Attribution map
Theoretical study and pipeline structure analysis of stable diffusion
Implementation of Diffusion Attentive Attribution Map (DAAM)
Designing additional experiments and analysis for meaningful interpretation of attribution map
Technical report writing
Post-hoc calibration using Ornstein-Uhlenbeck process @CAU
XAI Uncertinty calibration SDE Ornstein-Uhlenbeck process Temperature Scaling
Team research project in Explainable Artificial Intelligence 22f
Sep. 2022 ~ Dec. 2022
Evaluation on experiment method and proposed method
Experiment using synthetic data and visualization
Report writing
Highway Drone Control Automation @NextLab
OpticalFlow DeepSort EfficientNet HoughTransform CannyEdgeDetection FeatureDescriptor
Automate existing drone control work in highway, recognizing first lane region and vehicle type
Nov. 2021 ~ Dec. 2021
development of first lane recognition
make a pipeline between lane recognition module and vehicle type recognition
make a real demo at cheon-an highway
Subtitle Anomaly Detection @NextLab
EasyOCR CRAFT multi-threading ONNX
Create an algorithm checking subtitle anomaly in streaming videos
Sep. 2021
create subtitle datasets
develop a subtitle detection & recognition module
Implementation of whole pipeline including multi-processing
P&ID (Piping and Instrumentation Diagram) @JBNU
Recognize symbols and text of existing factory drawings and digitalize them
Apr. 2019 ~ Aug.2021
P&ID TesseractOCR EasyOCR CLEval MVVM
text detection & recognition
test and study various OCR models
create a pre, post processing algorithm to improve performance
make better OCR performance
make GUI to visualize and modify drawing recognition results
publication : [pdf]
Handwriting & News PDF OCR @NextLab
CRAFT deep-text-recognition-benchmark CTC ResNet EasyOCR VGGNet LSTM
OCR with text image in PDF
Oct. 2021 ~ Dec. 2021
Implementation of End-to-End OCR pipeline using the model structure
various experiment with data augmentation and pre-, post-processing
make better performance than original EasyOCR in Korean recognition
Colleful @JBNU
Java Spring Boot git
Development of meeting matching platform
Aug. 2020 ~ Nov. 2020
Develop back-end with java spring boot
Implementation of team matching API
Implementation of User management API for admin
Develop a department info crawler
Experience development rules and collaborative culture
Contribution to EasyOCR @JBNU
EasyOCR git
Contribution to a popular OCR framework contribution
Aug. 2020 ~ Nov.2020
solve a frequently uploaded issue
add a feature which makes possible to recognize rotated text
fix an error about post processing logic
Mask Detector @JBNU
Raspberry pi Embedded system MobileNet
Mask Detector against Covid-19 using Raspberry pi 4
Jun. 2021
transfer learning with Mobile Net V2 using Kaggle’s mask dataset
Implementation of running buzzer, camera, LED module