Continual multimodal forgery detection improves with stable authentic referencing
ARISE: Authentic Reference-Informed Stability Enhancement for Continual Multimodal Forgery Detection
Computer Vision and Pattern Recognition
Summary
Fake images and texts keep evolving, making it hard for detection systems trained on old fakes to spot new ones. The paper introduces a way to remember and learn from past real content without forgetting it, while adapting to new types of forgeries. The authors create a system that uses real examples as a stable reference and selectively focuses training on the parts that are most vulnerable to forgetting. Their tests show better long-term performance compared to other learning strategies.
What this means in practice
- •For security teams: Maintain effective detection of new multimedia fakes by continuously updating detectors without losing accuracy on known attacks.
- •For digital media platforms: Improve reliability of automated real/fake content flagging by integrating stable references from authentic media to handle evolving forgery methods.
- •For cybersecurity product developers: Design more robust forgery detection tools that adapt to new forgeries while preserving past knowledge using replay-informed continual learning techniques.$Commercial implications: Enables creation of advanced forgery detection software that stays effective over time for client cybersecurity solutions.
Authors
Yuchen Zhang, Yaxiong Wang, Yujiao Wu, Lianwei Wu, Qi Zhang, Li Zhu, Chaofeng Dong, Le Wu, Zhedong Zheng
Abstract
Multimodal forgery detection is commonly studied under a static assumption: training attacks are expected to approximate those encountered after deployment. This assumption becomes increasingly fragile as generators, editing systems, and multimodal large language models continually introduce new forgery mechanisms and forensic evidence. We formulate this problem as attack-incremental continual multimodal forgery detection, where the binary real/fake interface remains fixed while emerging attacks must be learned without forgetting previous ones. We introduce Fable-Bench, which organizes attack evolution through complementary joint-modality, text-side, and image-side protocols. We further propose ARISE (Authentic Reference-Informed Stability Enhancement), motivated by the open-ended forged class and the comparatively reusable geometry of authentic content. ARISE fits a real-only Gaussian mixture model to projected authentic replay representations as a compact historical reference. Meanwhile, the proposed Gaussian-Responsibility-Aware Stability Preservation (GRASP) derives a sample-adaptive anisotropic metric that strongly constrains historically stable representation directions while retaining plasticity elsewhere. Our Vulnerability-Aware Replay (VAR) further estimates stage-wise vulnerability from authentic-density degradation, forged-to-authentic boundary confusion, and authentic-mode occupancy drift, then assigns bounded replay emphasis to fragile stages. All adaptive signals are obtained from training-time replay statistics, while inference uses only the original multimodal backbone and binary classification head. Experiments across three attack-evolution protocols and multiple detector architectures demonstrate improved historical retention and final average performance over representative continual-learning baselines.