Deep Learning and Mathematical Methods for Deepfake Detection

The rapid progress of generative models has significantly increased the realism and diversity of synthetic media, challenging the reliability of digital information. Deepfake detection and image source attribution have therefore become central research problems. Deep learning has led major advances through powerful data-driven models capable of capturing subtle artifacts and complex distributional patterns. However, many approaches rely on heavily supervised training and substantial computational resources, raising concerns about scalability and robustness.

Recent work has explored detection and attribution strategies based on the internal representations of large foundation models, such as diffusion models and vision transformers. At the same time, mathematically grounded methods—from information theory to geometry, topology, and spectral analysis—offer principled tools to characterize structural differences between real and generated data.

About the workshop

This workshop brings together researchers from deep learning, applied mathematics, and computer vision to explore hybrid approaches for deepfake detection and attribution that integrate modern generative models with mathematical methods such as information theory, topology, geometry, and spectral analysis. 

Moreover, a challenge will be held on the rapidly evolving task of Synthetic Image Source Attribution. The challenge will be carried out on an attribution dataset, comprising state-of-the-art text-to-image generators. 

🚨Registrations and Submissions are now open!