11 papers
Where Identity Lives: Localized, Retain-Free Identity Unlearning in Multimodal Large Language Models
Kangwook Ko, Jaehyuk Jang, Wonjun Lee +2
Removing a specific individual's information from multimodal large language models (MLLMs) is often needed after deployment, but existing methods rely on a retain set, which is har…
AIM: Anchor Identity Features, Then Match for Multimodal Large Language Model Unlearning
Wonjun Lee, Jaehyuk Jang, Kangwook Ko +2
Multimodal large language models (MLLMs) can memorize identity-specific facts about people in their fine-tuning data, creating privacy risks when a person requests deletion. Existi…
Safety in Batches? Understanding and Mitigating Safety Failures in Batch Prompting
Kihyun Kim, Hee-Seon Kim, Wonjun Lee +1
Batch prompting is a practical inference strategy for large language models, but its safety implications remain underexplored. We show that the success of batch prompting for utili…
SubT: Subspace Tuning for Few-shot Generalization of Audio-Language Models
Jaehyuk Jang, Kangwook Ko, Wonjun Lee +1
Few-shot parameter-efficient adaptation of pretrained Audio--Language Models (ALMs) often improves seen-class performance at the cost of unseen-class generalization, leading to the…
Jailbreak to Protect: Buffering and Reinforcing via Temporary Jailbreaking for Safe Fine-Tuning in Large Language Models
Seokil Ham, Jaehyuk Jang, Wonjun Lee +1
Fine-tuning-as-a-Service (FaaS) enables personalization of large language models (LLMs), but it can weaken safety-alignment under harmful fine-tuning attacks. Recent work has shown…
StreamTTO: Efficient Online Test-Time Optimization for Video Depth Completion
Minseok Seo, Wonjun Lee, Jaehyuk Jang +1
Monocular depth foundation models generalize across diverse scenes, but recovering accurate metric depth consistent with a target sensor remains challenging under sensor variation…
Generalizable Prompt Tuning for Audio-Language Models via Semantic Expansion
Jaehyuk Jang, Wonjun Lee, Kangwook Ko +1
Prompt tuning has achieved remarkable progress in vision-language models (VLMs) and is recently being adopted for audio-language models (ALMs). However, its generalization ability…
SELFI: Selective Fusion of Identity for Generalizable Deepfake Detection
Younghun Kim, Minsuk Jang, Myung-Joon Kwon +2
Face identity provides a powerful signal for deepfake detection. Prior studies show that even when not explicitly modeled, classifiers often learn identity features implicitly. Thi…
Benign-to-Toxic Jailbreaking: Inducing Harmful Responses from Harmless Prompts
Hee-Seon Kim, Minbeom Kim, Wonjun Lee +2
Optimization-based jailbreaks typically adopt the Toxic-Continuation setting in large vision-language models (LVLMs), following the standard next-token prediction objective. In thi…
FRIDAY: Mitigating Unintentional Facial Identity in Deepfake Detectors Guided by Facial Recognizers
Younhun Kim, Myung-Joon Kwon, Wonjun Lee +1
Previous Deepfake detection methods perform well within their training domains, but their effectiveness diminishes significantly with new synthesis techniques. Recent studies have…
SAFIRE: Segment Any Forged Image Region
Myung-Joon Kwon, Wonjun Lee, Seung-Hun Nam +2
Most techniques approach the problem of image forgery localization as a binary segmentation task, training neural networks to label original areas as 0 and forged areas as 1. In co…