중앙대학교 Chung-Ang University

Trustworthy
Artificial Intelligence LabTrustworthy AI Lab

신뢰할 수 있는 인공지능 연구실 Hoki Kim

01

Mission

믿을 수 있는 인공지능을 위한
세 가지 핵심 개념 연구

Researching three core concepts
for trustworthy AI

AI Security

AI 보안

악의적 조작과 공격으로부터 AI를 보호하고, LLM·AI 에이전트 악용 위협까지 탐지·방어합니다.

적대적 공격 탈옥 AI 해킹

AI Security

We protect AI from malicious manipulation and attacks, and detect and defend against threats that abuse LLMs and AI agents.

Adversarial Attack Jailbreak AI Hacking
Learn more

AI Privacy

AI 프라이버시

민감정보를 안전하게 학습에 활용하고, AI를 통한 유출을 차단해 개인정보 보호와 규제 준수를 강화합니다.

머신언러닝 개인정보 추출 공격

AI Privacy

We make sensitive data safe to train on and block leakage through AI, strengthening personal data protection and regulatory compliance.

Machine Unlearning Privacy Extraction
Learn more

AI Explainability

AI 설명성

AI의 작동 과정과 의사결정 근거를 명확히 설명해 의사결정의 투명성과 검증 가능성을 높입니다.

이상탐지 결함진단

AI Explainability

We clearly explain how AI works and the rationale behind its decisions, raising the transparency and verifiability of those decisions.

Anomaly Detection Fault Diagnosis
Learn more
02

Collaboration

다양한 파트너와의
신뢰할 수 있는 결과

Trustworthy results
with diverse partners

NAVER Cloud
보안 특화 AI 파운데이션 모델 개발 Cybersecurity-Specialized AI Foundation Model
네이버클라우드 NAVER Cloud
LG CNS
보안 특화 AI 파운데이션 모델 개발 Cybersecurity-Specialized AI Foundation Model
LG CNS LG CNS
Financial Security Institute
설명 가능한 자금세탁방지 AI 개발 Explainable Anti-Money-Laundering AI
금융보안원 Financial Security Institute
KISA
「AI 보안 위협 대응 매뉴얼」 제작 AI Security Threat Mitigation Manual
KISA KISA
SK Shieldus
「AI 보안 위협 대응 매뉴얼」 제작 AI Security Threat Mitigation Manual
SK쉴더스 SK Shieldus
KIAT
온디바이스 AI 보안 기술 협력 On-device AI Security Technology
KIAT KIAT
TAP
온디바이스 AI 보안 기술 협력 On-device AI Security Technology
탑전자산업 TAP
UROCK
디지털 포렌식 AI 기술 협력 Digital Forensic AI Technology
유락 UROCK
Coxwave
AI 에이전트 페르소나 신뢰성 연구 AI Agent Persona Trustworthiness
콕스웨이브 Coxwave
MSIT
규제 준수형 언러닝 기술 개발 Regulatory-Compliant Unlearning
과학기술정보통신부 MSIT
NICE Info
기술특례상장 자문위원 AI Evaluation Committee
나이스 평가정보 NICE Info
KimCaddie
골프 스윙 분석 AI 기술 협력 Golf Swing Analysis AI Technology
김캐디 KimCaddie
03

Research

최근 5년 최상위 AI 학회 & SCI 논문
25편 이상 게재

25+ papers in top AI conferences & SCI journals
over the past 5 years

  1. EMNLP
    EMNLP2026

    Exponential-Tilted Unlearning
  2. ICML
    ICML2026

    Proactive Defense Benchmark against Deepfake Generation
  3. ICML
    ICML2026

    Position: Current Model Cards Are Insufficient for Downstream Governance of Open-Weight Foundation Models
  4. ICML
    ICML2026

    Stability Analysis of Sharpness-Aware Minimization
  5. ICLR
    ICLR2026

    Co-occurring Associated REtained concepts in Diffusion Unlearning
  6. NeurIPS
    NeurIPS2025

    Unlearning-Aware Minimization
  7. NeurIPS
    NeurIPS2024

    Are Self-Attentions Effective for Time Series Forecasting?
  8. AAAI
    AAAI2024

    Fair Sampling in Diffusion Models through Switching Mechanism
  9. NeurIPS
    NeurIPS2023

    Fantastic Robustness Measures: The Secrets of Robust Generalization
  10. ICML
    ICML2023

    Differentially Private Sharpness-Aware Training
  11. AAAI
    AAAI2021

    Understanding Catastrophic Overfitting in Single-step Adversarial Training
  1. Elsevier
    Top 5%2027

    Toward Regulation-Friendly Bearing Fault Diagnosis via Structural Embedding Preserving Machine Unlearning
  2. IEEE
    Top 5%2026

    Adversarial Retain-Free Unlearning for Bearing Prognostics and Health Management
  3. Springer
    Top 1%2026

    Pattern-guided forecasting framework for metal price prediction with grouping decomposed series
  4. Elsevier
    Top 10%2025

    Towards undetectable adversarial attack on time series classification
  5. Elsevier
    Top 5%2024

    Evaluating practical adversarial robustness of fault diagnosis systems via spectrogram-aware ensemble method
  6. IEEE
    Top 1%2022

    Graddiv: Adversarial Robustness of Randomized Neural Networks via Gradient Diversity Regularization
Google Scholar GitHub torchattacks 2.2k torchbnn 560
04

History

2026 13건 13 entries

09.17

연구실 논문 “Toward Regulation-Friendly Bearing Fault Diagnosis via Structural Embedding Preserving Machine Unlearning”이 Advanced Engineering Informatics에 게재되었습니다.

Our paper “Toward Regulation-Friendly Bearing Fault Diagnosis via Structural Embedding Preserving Machine Unlearning” is published in Advanced Engineering Informatics.

09.17

연구실 논문 “Exponential-Tilted Unlearning”이 EMNLP 2026 Findings에 채택되었습니다.

Our paper “Exponential-Tilted Unlearning” is accepted to EMNLP 2026 Findings.

09.17

김호기 교수가 ICLR 2027의 Area Chair로 선임되었습니다.

Prof. Hoki Kim has been appointed as an Area Chair for ICLR 2027.

2026.09.03NEWS

본 연구실이 참여한 네이버클라우드 컨소시엄이 과학기술정보통신부 「사이버 보안 특화 AI 파운데이션 모델 개발」 사업 수행기관으로 선정되었습니다.

The NAVER Cloud consortium that our lab participates in was selected as the executing organization for MSIT’s Cybersecurity-Specialized AI Foundation Model Development project.

2026.07.08NEWS

정보보호의 날을 맞아, 저희 연구실이 KISA·SK쉴더스와 공동 제작한 「AI 보안 위협 대응 매뉴얼」이 배포되었습니다.

The AI Security Threat Mitigation Manual, which our lab co-developed with KISA and SK Shieldus, was released for Information Security Day.

2026.05.08NEWS

연구실 논문 3편이 ICML 2026에 동시 채택되었습니다.

Three of our lab’s papers were accepted to ICML 2026.

05.01

연구실 논문 “Position: Current Model Cards Are Insufficient for Downstream Governance of Open-Weight Foundation Models”이 ICML 2026에 채택되었습니다.

Our paper “Position: Current Model Cards Are Insufficient for Downstream Governance of Open-Weight Foundation Models” is accepted to ICML 2026.

05.01

연구실 논문 “Proactive Defense Benchmark against Deepfake Generation”이 ICML 2026에 채택되었습니다.

Our paper “Proactive Defense Benchmark against Deepfake Generation” is accepted to ICML 2026.

05.01

연구실 논문 “Stability Analysis of Sharpness-Aware Minimization”이 ICML 2026에 채택되었습니다.

Our paper “Stability Analysis of Sharpness-Aware Minimization” is accepted to ICML 2026.

2026.03.11NEWS

연구실 논문 “Adversarial Retain-Free Unlearning for Bearing Prognostics and Health Management”이 IEEE Transactions on Industrial Informatics에 게재되었습니다.

Our paper “Adversarial Retain-Free Unlearning for Bearing Prognostics and Health Management” is published in IEEE Transactions on Industrial Informatics.

03.01

김호기 교수가 중앙대학교 보안대학원 개인정보보호학과 학과장으로 부임하였습니다.

Prof. Hoki Kim has been appointed as Department Chair of the Department of Privacy Protection, Graduate School of Security, Chung-Ang University.

01.30

연구실 논문 “Pattern-guided forecasting framework for metal price prediction with grouping decomposed series”가 Financial Innovation에 게재되었습니다.

Our paper “Pattern-guided forecasting framework for metal price prediction with grouping decomposed series” is published in Financial Innovation.

01.26

연구실 논문 “Co-occurring Associated REtained concepts in Diffusion Unlearning”이 ICLR 2026에 채택되었습니다.

Our paper “Co-occurring Associated REtained concepts in Diffusion Unlearning” is accepted to ICLR 2026.

2025 5건 5 entries

12.04

NeurIPS 2025에서 포스터 발표를 위해 미국 샌디에고 방문 예정입니다.

Attending NeurIPS 2025 for poster presentation in San Diego, USA.

11.29

최대승, 김재성, 김민철 연구원이 “한국전자거래학회 2025 추계학술대회”에서 우수 논문상을 수상하였습니다.

Research interns Daeseung Choi, Jaeseong Kim, and Mincheol Kim received the Outstanding Paper Award at the 2025 Fall Conference of the Korea Electronic Commerce Association.

2025.11.18NEWS

김수민, 장희원, 전소현 연구원이 “중앙대학교 AI·SW융합 우수성과 발표대회”에서 최우수상을 수상하였습니다.

Research interns Sumin Kim, Heewon Jang, and Sohyun Jeon received the Grand Prize at the Chung-Ang University AI·SW Convergence Excellence Presentation Competition.

2025.09.18NEWS

연구실 논문 “Unlearning Aware Minimization”이 NeurIPS 2025에 채택되었습니다.

Our paper “Unlearning Aware Minimization” is accepted to NeurIPS 2025.

06.05

연구실 논문 “CaddieSet: A Golf Swing Dataset with Human Joint Features and Ball Information”이 11th International Workshop on Computer Vision in Sports (CVsports) at CVPR 2025에 게재되었습니다.

Our paper “CaddieSet: A Golf Swing Dataset with Human Joint Features and Ball Information” is published in the 11th International Workshop on Computer Vision in Sports (CVsports) at CVPR 2025.

2024 4건 4 entries

12.07

NeurIPS 2024에서 포스터 발표를 위해 캐나다 밴쿠버 🇨🇦 방문 예정입니다.

Attending NeurIPS 2024 for poster presentation in Vancouver, Canada 🇨🇦.

09.26

연구실 논문 “Are Self-Attentions Effective for Time Series Forecasting”이 NeurIPS 2024에 채택되었습니다.

Our paper “Are Self-Attentions Effective for Time Series Forecasting” is accepted to NeurIPS 2024.

04.01

연구실 논문 “Evaluating practical adversarial robustness of fault diagnosis systems via spectrogram-aware ensemble method”가 Engineering Applications of Artificial Intelligence에 게재되었습니다.

Our paper “Evaluating practical adversarial robustness of fault diagnosis systems via spectrogram-aware ensemble method” is published in Engineering Applications of Artificial Intelligence.

03.24

연구실 논문 “Fair sampling in diffusion models through switching mechanism”이 AAAI 2024에 채택되었습니다.

Our paper “Fair sampling in diffusion models through switching mechanism” is accepted to AAAI 2024.

2023 4건 4 entries

10.01

연구실 논문 “Bridged adversarial training”가 Neural Networks에 게재되었습니다.

Our paper “Bridged adversarial training” is published in Neural Networks.

09.01

연구실 논문 “Fast sharpness-aware training for periodic time series classification and forecasting”가 Applied Soft Computing에 게재되었습니다.

Our paper “Fast sharpness-aware training for periodic time series classification and forecasting” is published in Applied Soft Computing.

06.09

연구실 논문 “Fantastic robustness measures the secrets of robust generalization”이 NeurIPS 2023에 채택되었습니다.

Our paper “Fantastic robustness measures the secrets of robust generalization” is accepted to NeurIPS 2023.

05.09

연구실 논문 “Differentially Private Sharpness-Aware Training”이 ICML 2023에 채택되었습니다.

Our paper “Differentially Private Sharpness-Aware Training” is accepted to ICML 2023.

TrustworthyAI

Location

중앙대학교 서울캠퍼스 310관 1006호
Bldg. 310, Rm. 1006, Chung-Ang Univ. Seoul Campus

Contact

hokikim@cau.ac.kr