Schedule

Conference Venue: The IEEE CAI 2025 will be held at the prestigious HYATT REGENCY SANTA CLARA hotel. Conveniently located in the heart of Silicon Valley, this elegant venue is directly connected to the Santa Clara Convention Center and steps away from Levi’s Stadium. For more details about the hotel and its amenities, visit https://www.hyatt.com/en-US/group-booking/CLARA/G-EEEI 

Program: https://cai.ieee.org/2025/program2025/ 

Tamar' Talk:

Title: Modular Minds: Rethinking Deepfake Detection Through Model Disentanglement

Abstract: As AI-generated content becomes increasingly indistinguishable from human-created media, traditional detection approaches face significant challenges. This talk explores the evolution of deepfake detection from artifact hunting to a more fundamental rethinking of the problem. The core thesis proposes a paradigm shift: rather than focusing on task-specific detection endpoints, we should prioritize the underlying representations and model disentanglement. By extracting and isolating information in projected latent spaces, we can develop more generalizable, robust, and transferable solutions that transcend specific detection tasks. This approach represents not only a technical advancement but a conceptual shift in how we understand and address the broader implications of synthetic media. The talk will conclude with reflections on future research directions and the broader impact of representation-focused approaches across multiple domains.

Tal's Talk:

Title: Deep fake detection: A (deep) Fake Problem? 

Abstract: Generative AI is transforming online media by enabling the creation of highly convincing synthetic content. The growing prevalence of AI-generated content introduces new risks to privacy and misinformation, while traditional detection methods are rapidly becoming insufficient or obsolete. In this talk, I will present two approaches that look beyond the conventional aim of detecting fakes. The first is model attribution, where I will present a "model parsing" technique that reverse-engineers generative models from their image outputs, making it possible to trace AI-generated media back to its source and expose coordinated misinformation campaigns. The second is media provenance, where I will present work on proactive methods that embed protective watermarks into images, enabling the detection of unauthorized manipulations while preserving owners' privacy and intellectual property rights. While the challenges of generative AI will persist, these and similar methods offer promising frameworks for addressing evolving threats, helping preserve the trustworthiness of digital media in an increasingly AI-driven world.