analiza aiqbench
This arXiv preprint tests whether federated training, which keeps raw data local, undermines Grad-CAM explanations for an EfficientNet-B4 deepfake detector on FaceForensics++. Its main finding is non-monotonic: federated models match or beat a centralized baseline at near-IID and moderate heterogeneity, but explanations become spatially incoherent and causally misleading under extreme client heterogeneity. Readers deploying privacy-preserving detection should treat the heterogeneity threshold as the key variable to measure, since it determines when saliency maps can be trusted in forensic review.
Does Federated Training Degrade Explainability? A Comparative Analysis of Grad-CAM Faithfulness in Centralized vs. Federated Deepfake Detection
arXiv:2610.39483v1 Announce Type: new Abstract: Automated detection has become the essential safeguard against realistic synthetic media, but deploying it at scale requires training on sensitive data from multiple organizations, putting privacy and the transparency demanded by forensic review in direct conflict. Federated learning resolves the privacy side by sharing only model updates rather than raw data, but whether this comes at the cost of trustworthy gradient-based explanations, which analysts rely on to decide whether or not to trust a flagged clip, has not been studied. We present the systematic study of Grad-CAM explanation faithfulness under federated deepfake detection, evaluating an EfficientNet-B4 detector on FaceForensics++ across five federated clients under varying degrees of data heterogeneity. We find that the relationship between federation and explanation quality is non-monotonic: at near-IID (Independent and Identically Distributed Data) and moderate heterogeneity, federated models match or exceed the faithfulness of a centralized baseline, but at extreme heterogeneity, explanations break down entirely, becoming spatially incoherent and, on average, causally misleading. This degradation is not a general cost of federated training but a threshold effect tied to client heterogeneity, and we offer practical guidance for identifying when privacy-preserving deepfake detection can be deployed safely and when its explanations can no longer be trusted. Code and data partitions are available for reproducibility.
Title:Does Federated Training Degrade Explainability? A Comparative Analysis of Grad-CAM Faithfulness in Centralized vs. Federated Deepfake Detection
Abstract:Automated detection has become the essential safeguard against realistic synthetic media, but deploying it at scale requires training on sensitive data from multiple organizations, putting privacy and the transparency demanded by forensic review in direct conflict. Federated learning resolves the privacy side by sharing only model updates rather than raw data, but whether this comes at the cost of trustworthy gradient-based explanations, which analysts rely on to decide whether or not to trust a flagged clip, has not been studied. We present the systematic study of Grad-CAM explanation faithfulness under federated deepfake detection, evaluating an EfficientNet-B4 detector on FaceForensics++ across five federated clients under varying degrees of data heterogeneity. We find that the relationship between federation and explanation quality is non-monotonic: at near-IID (Independent and Identically Distributed Data) and moderate heterogeneity, federated models match or exceed the faithfulness of a centralized baseline, but at extreme heterogeneity, explanations break down entirely, becoming spatially incoherent and, on average, causally misleading. This degradation is not a general cost of federated training but a threshold effect tied to client heterogeneity, and we offer practical guidance for identifying when privacy-preserving deepfake detection can be deployed safely and when its explanations can no longer be trusted. Code and data partitions are available for reproducibility.
Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2610.39483 [cs.AI]
(or arXiv:2610.39483v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.39483 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 330 cuvinte
Subiecte conexe
Listening with Attention: Entropy-Guided Explainability for Transformer-Based Audio Models
arXiv · 15 iun. 2026
Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability
arXiv · 27 oct. 2026
Efficient Reasoning Training Does Not Always Harm CoT Faithfulness and Monitorability
arXiv · 6 oct. 2026
I Don't Miss You, but I Do: Self-Explanation Faithfulness of Modality Missingness in Vision-Language Models
arXiv · 9 sept. 2026