-
Preprint, 2026
Shunchang Liu,
Lukas Fluri,
Xin Chen,
Francesco Croce
Shows that narrow multimodal fine-tuning induces unsafe behavior on unrelated tasks, extending emergent misalignment from text-only to multimodal models.
PaperCode
-
Preprint, 2026
Zhuan Shi,
Shunchang Liu,
Alireza Dehghanpour Farashah,
Qian Yang,
Han Yu,
Cao Yang,
Chaochao Chen,
Yuping Yan,
Yaochu Jin,
Golnoosh Farnadi,
Lingjuan Lyu
Presents a taxonomy for IP protection in visual generative models that classifies methods into Information Exposure Control, Generative Behavior Constraint, and Attribution & Accountability.
Paper
-
Developments in the Built Environment, 2026
Shuting Mi,
Xiaofei Wu,
Shunchang Liu,
Zhao Huang,
Catherine De Wolf,
Mennatallah El-Assady,
Pei-Yu Wu
Proposes a hybrid spatial-semantic modeling framework for automating building product inventories.
Paper
-
ICML @CompLearn Workshop & @Mech Interp Workshop, 2026
Shunchang Liu,
Xin Chen,
Belen Martin Urcelay,
Francesco Croce
Finds that RLHF reward models can reverse preferences under semantic-preserving perturbations, traces this instability to sparse internal features, and mitigates it without retraining the model.
PaperCode
-
NeurIPS @GenAI4Health Workshop, 2025
Jianzhou Yao*,
Shunchang Liu*,
Guillaume Drui,
Rikard Pettersson,
Alessandro Blasimme,
Sara Kijewski
(* equal contribution)
Evaluates LLMs in medical diagnostic scenarios and finds that they adapt explanations to socio-demographic variables but generate overly complex content and display biased affective empathy.
PaperCode
-
ACM International Conference on Multimedia (MM), 2025
Shunchang Liu*,
Zhuan Shi*,
Lingjuan Lyu,
Yaochu Jin,
Boi Faltings
(* equal contribution)
Builds CopyJudge, a VLM framework that assesses copyright infringement through multi-agent debate and reduces infringement risk in diffusion models through text-prompt and latent-noise optimization.
PaperCodeVideo
Thanks for Anderson's insightful commentary and sharing (1, 2) by the Korea Copyright Commission.
-
CVPR @AdvCV Workshop, 2023
Yisong Xiao,
Tianyuan Zhang,
Shunchang Liu,
Haotong Qin
Evaluates the robustness of quantized models on ImageNet, showing that lower-bit quantization is more resilient to adversarial attacks but more susceptible to natural corruptions and systematic noises.
Paper
-
CVPR @AdvCV Workshop, 2023
Tony Ma,
Songze Li,
Yisong Xiao,
Shunchang Liu#
(# corresponding author)
Proposes VRAP, a visual relation-based cross-task adversarial patch method that uses scene graphs to disrupt visual reasoning tasks such as visual question answering and image captioning.
Paper
-
ACM SIGSAC Conference on Computer and Communications Security (CCS), 2022
Shunchang Liu*,
Jiakai Wang*,
Aishan Liu,
Yingwei Li,
Yijie Gao,
Xianglong Liu,
Dacheng Tao
(* equal contribution)
Designs scale- and position-aware adversarial patches that achieve SOTA attacks on crowd-counting models, improving robustness to complex backgrounds and cross-dataset generalization through adversarial training.
CodePaperVideo
-
Neurocomputing, 2022
Aishan Liu,
Huiyuan Xie,
Xianglong Liu,
Zixin Yin,
Shunchang Liu
Revisits the AVSD task and shows biases in models, datasets, and evaluation metrics.
Paper
-
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021 Oral
Jiakai Wang,
Aishan Liu,
Zixin Yin,
Shunchang Liu,
Shiyu Tang,
Xianglong Liu
Proposes the Dual Attention Suppression (DAS) attack, which generates visually-natural physical adversarial camouflages with strong transferability by suppressing both model and human attention.
PaperCode
No papers match this filter.
