ML Security PapersFrom Spurious to Causal: Low-rank Orthogonal Subspace Intervention for Generalizable Face Forgery Detection
Chi Wang , Xinjue Hu , Boyu Wang , Ziwen He , Zhangjie Fu
Published on arXiv
2601.11915
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Achieves state-of-the-art cross-dataset generalization on face forgery detection benchmarks using only 0.43M trainable parameters by removing spurious correlation subspaces from feature representations.
Low-rank Orthogonal Subspace Intervention (LOSI)
Novel technique introduced
The generalization problem remains a critical challenge in face forgery detection. Some researches have discovered that ``a backdoor path" in the representations from forgery-irrelevant information to labels induces biased learning, thereby hindering the generalization. In this paper, these forgery-irrelevant information are collectively termed spurious correlations factors. Previous methods predominantly focused on identifying concrete, specific spurious correlation and designing corresponding solutions to address them. However, spurious correlations arise from unobservable confounding factors, making it impractical to identify and address each one individually. To address this, we propose an intervention paradigm for representation space. Instead of tracking and blocking various instance-level spurious correlation one by one, we uniformly model them as a low-rank subspace and intervene in them. Specifically, we decompose spurious correlation features into a low-rank subspace via orthogonal low-rank projection, subsequently removing this subspace from the original representation and training its orthogonal complement to capture forgery-related features. This low-rank projection removal effectively eliminates spurious correlation factors, ensuring that classification decision is based on authentic forgery cues. With only 0.43M trainable parameters, our method achieves state-of-the-art performance across several benchmarks, demonstrating excellent robustness and generalization.
Key Contributions
- Proposes a low-rank orthogonal subspace intervention paradigm that uniformly models all spurious correlation factors as a low-rank subspace and removes them from representations, rather than addressing each spurious factor individually.
- Decomposes forgery-irrelevant features via orthogonal low-rank projection and trains only the orthogonal complement to capture causal forgery cues.
- Achieves state-of-the-art generalization across multiple face forgery detection benchmarks with only 0.43M trainable parameters.
🛡️ Threat Analysis
Proposes a novel face forgery detection architecture that addresses a core forensic challenge (spurious correlations degrading generalization). Deepfake/face forgery detection is explicitly listed under ML09 AI-generated content detection, and the paper's primary contribution is a new detection technique, not a domain application of existing methods.