Sungmin Han

PhD student, School of Cybersecurity, Korea University

Joined

May 2023

Names

Sungmin Han (Preferred)

Emails

****@korea.ac.kr (Confirmed)

Personal Links

Career & Education History

PhD student

School of Cybersecurity, Korea University (korea.ac.kr)

2024 – Present

MS student

School of Cybersecurity, Korea University (korea.ac.kr)

2021 – 2024

Undergrad student

Kwangwoon University (kw.ac.kr)

2015 – 2021

Advisors, Relations & Conflicts

PhD Advisor

Sangkyun Lee

2024 – Present

Coauthor

Jeonghyun Lee

2019 – Present

Expertise

eXplainable Artificial Intelligence

2021 – Present

Trustworthy AI

2021 – Present

Publications

Contrast-CAT: Contrasting Activations for Enhanced Interpretability in Transformer-based Text Classifiers
Sungmin Han, Jeonghyun Lee, Sangkyun Lee
- UAI 2025 Poster
- Readers: Everyone
Improving Transformer Interpretability with Activation Contrast-Based Attribution
Sungmin Han, Jeonghyun Lee, Sangkyun Lee
- Submitted to ICLR 2025
- Readers: Everyone
Cluster-Driven Adversarial Perturbations for Robust Contrastive Learning
Seongyun Seo, Sungmin Han, Jeonghyun Lee, Sangkyun Lee
- Submitted to ICLR 2025
- Readers: Everyone
Model Stealing Defense against Exploiting Information Leak through the Interpretation of Deep Neural Nets
Jeonghyun Lee, Sungmin Han, Sangkyun Lee
- Published: 01 Jan 2022, Last Modified: 10 May 2023
- IJCAI 2022
- Readers: Everyone
Libra-CAM: An Activation-Based Attribution Based on the Linear Approximation of Deep Neural Nets and Threshold Calibration
Sangkyun Lee, Sungmin Han
- Published: 01 Jan 2022, Last Modified: 10 May 2023
- IJCAI 2022
- Readers: Everyone
Hunt for Unseen Intrusion: Multi-Head Self-Attention Neural Detector
Seongyun Seo, Sungmin Han, Janghyeon Park, Shin-Woo Shim, Han-Eul Ryu, Byoungmo Cho, Sangkyun Lee
- Published: 01 Jan 2021, Last Modified: 10 May 2023
- IEEE Access 2021
- Readers: Everyone

Sungmin Han

Names

Emails

Personal Links

Career & Education History

Advisors, Relations & Conflicts

Expertise

Publications

Contrast-CAT: Contrasting Activations for Enhanced Interpretability in Transformer-based Text Classifiers

Improving Transformer Interpretability with Activation Contrast-Based Attribution

Cluster-Driven Adversarial Perturbations for Robust Contrastive Learning

Model Stealing Defense against Exploiting Information Leak through the Interpretation of Deep Neural Nets

Libra-CAM: An Activation-Based Attribution Based on the Linear Approximation of Deep Neural Nets and Threshold Calibration

Hunt for Unseen Intrusion: Multi-Head Self-Attention Neural Detector

Co-Authors