Investigating audio-visual representations for deepfake detection

Bucharest Deep Learning is back with another exciting session! Join us as we delve into the frontline of multimedia forensics alongside Dragos Alexandru Boldisor, presenting his recently accepted CVPR 2026 paper on audio-visual deepfake detection.

The Talk: Investigating Self-Supervised Representations for Audio-Visual Deepfake Detection

Generative models are now capable of producing text, images, audio, and video that rival human creations, making reliable deepfake detection an urgent necessity. While self-supervised learning representations have shown great promise in this context, their specific potential for audio-visual deepfake detection remains underexplored. Often, these backbone features are used in isolation or buried within complex architectures, leaving it unclear exactly how they contribute to detection performance.

Join us to understand exactly what these self-supervised features capture, where their blind spots are, and how we can better leverage them to figure out what digital content we can actually trust.

Logistics:

• Date & Time: Tuesday, July 28 | 18:30 - 19:30
• Location: FMI New Building (Politehnica Business Tower)
• Address: Bulevardul Iuliu Maniu, nr. 15G, Etaj 5, Room 503