---
title: "CASIA's Direct Discrepancy Replay Beats Deepfake Detectors While Ditching Stored Faces"
subtitle: "A new continual learning approach for deepfake detection stores tiny distribution maps instead of face images, cutting memory 10x while beating prior state of the art."
url: "https://alphasignal.ai/news/casia-s-direct-discrepancy-replay-beats-deepfake-detectors-while-ditching"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-04-14T00:00:00.000Z"
updated: "2026-08-23T21:56:55.626Z"
type: "Paper"
topics:
  - "Image"
  - "Security"
subtopics:
  - "Image Generation"
  - "Red Teaming"
upvotes: 1088
source_url: "https://arxiv.org/abs/2604.12941"
comments: 0
truncated: true
---

# CASIA's Direct Discrepancy Replay Beats Deepfake Detectors While Ditching Stored Faces

> A new continual learning approach for deepfake detection stores tiny distribution maps instead of face images, cutting memory 10x while beating prior state of the art.

Deepfake generators keep multiplying, and every new family of forgeries poses the same headache for detectors: how do you teach a model the new manipulation without erasing what it learned about the old ones? A paper from researchers at CASIA and UCAS proposes a cleaner answer than the usual approaches, and it comes with a useful side effect for privacy.

The work introduces [Direct Discrepancy Replay](https://arxiv.org/abs/2604.12941), a continual learning framework for face forgery detection that stops storing past face images entirely. It condenses the statistical gap between real and fake faces into a tiny bank of learnable maps, then reconstructs training examples on the fly by combining those maps with real faces from the current task.

## Why detectors forget

Continual Face Forgery Detection (CFFD) is the task of updating a detector as new manipulation techniques appear. Continual learning has become the go-to paradigm here because it lets models acquire new knowledge while retaining previously learned capabilities, typically through sample replay, parameter regularization, or gradient projection. Rehearsing a few examples from earlier tasks during new training, known as replay, is the dominant strategy.

Both flavors of replay have real drawbacks:

* **Selection-based replay** keeps a small subset of historical images. Under tight memory budgets it misses the diversity of past forgery cues, and it retains actual identifiable faces in memory.
* **Perturbation-based replay**, like the [Historical Distribution Preserving (HDP)](https://link.springer.com/article/10.1007/s11263-024-02160-1) approach that uses universal adversarial perturbations to simulate past forgeries, avoids storing raw faces but stays tied to the previous decision boundary and drifts as the backbone updates.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [Direct Discrepancy Replay: Distribution-Discrepancy Condensation and Manifold-Consistent Replay for Continual Face Forgery Detection](https://arxiv.org/abs/2604.12941)
- [Continual Face Forgery Detection via Historical Distribution Preserving (HDP)](https://link.springer.com/article/10.1007/s11263-024-02160-1)
- [Face-D2CL: Multi-Domain Synergistic Representation with Dual Continual Learning for Facial DeepFake Detection](https://arxiv.org/html/2604.08159)
- [Unifying Locality of KANs and Feature Drift Compensation Projection for Data-free Replay based Continual Face Forgery Detection (KAN-CFD)](https://arxiv.org/abs/2508.03189</a>)