Suppressing Gradient Conflict for Generalizable Deepfake Detection

2025-07-29

AI Summary

In deepfake detection, joint training on original and online-synthesized forged images often induces gradient conflicts, degrading in-domain performance and impairing cross-domain generalization—contradicting the intuitive goal of “enhancing source-domain capability while improving efficiency.” This work is the first to formally characterize this gradient conflict mechanism and proposes a novel gradient-cooperative optimization framework: (1) Update Vector Search (UVS), which formulates gradient optimization as an extremum-seeking problem to explicitly suppress conflicting directions; and (2) Conflict-Gradient Reduction loss (CGR), which explicitly enforces a low-conflict feature space. Evaluated across multiple benchmarks, our method simultaneously boosts in-domain detection accuracy and cross-domain generalization, consistently outperforming state-of-the-art approaches by significant margins.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Robust deepfake detection models must be capable of generalizing to ever-evolving manipulation techniques beyond training data. A promising strategy is to augment the training data with online synthesized fake images containing broadly generalizable artifacts. However, in the context of deepfake detection, it is surprising that jointly training on both original and online synthesized forgeries may result in degraded performance. This contradicts the common belief that incorporating more source-domain data should enhance detection accuracy. Through empirical analysis, we trace this degradation to gradient conflicts during backpropagation which force a trade-off between source domain accuracy and target domain generalization. To overcome this issue, we propose a Conflict-Suppressed Deepfake Detection (CS-DFD) framework that explicitly mitigates the gradient conflict via two synergistic modules. First, an Update Vector Search (UVS) module searches for an alternative update vector near the initial gradient vector to reconcile the disparities of the original and online synthesized forgeries. By further transforming the search process into an extremum optimization problem, UVS yields the uniquely update vector, which maximizes the simultaneous loss reductions for each data type. Second, a Conflict Gradient Reduction (CGR) module enforces a low-conflict feature embedding space through a novel Conflict Descent Loss. This loss penalizes misaligned gradient directions and guides the learning of representations with aligned, non-conflicting gradients. The synergy of UVS and CGR alleviates gradient interference in both parameter optimization and representation learning. Experiments on multiple deepfake benchmarks demonstrate that CS-DFD achieves state-of-the-art performance in both in-domain detection accuracy and cross-domain generalization.

Problem

• Addressing gradient conflicts in deepfake detection training
• Improving generalization to unseen manipulation techniques
• Enhancing detection accuracy with conflict-suppressed learning

Innovation

• UVS module searches alternative update vector
• CGR module enforces low-conflict embedding space
• CS-DFD mitigates gradient conflict synergistically