Day 1 - Auguest 11 (9:45-11:45)
Talk: Multimedia Deepfake Detection (Remote Session)
Speaker: You (Neil) Zhang, Dolby, Menglu Li, Toronto Metropolitan University, Luchuan Song, University of Rochester
Abstract: The rapid advancement of deepfake technology, particularly influenced by the evolution of generative AI technology, is raising significant concerns across various sectors due to its potential misuse in creating deceptive audio-visual content. The potential misuse of deepfakes for creating deceptive content has prompted an urgent need for effective detection methodologies. Our proposed tutorial, "Multimedia Deepfake Detection," previously presented at ICME 2024, aims to address this challenge by convening experts from different but related research communities that focus on the current challenges brought about by deepfakes. The primary objective is to foster cross-disciplinary collaboration, exchanging insights and methodologies to enhance the effectiveness of audio-visual deepfake detection techniques.
The tutorial seeks to provide a comprehensive platform for researchers, practitioners, and enthusiasts to gain deep insights into the state-of-the-art techniques for detecting deepfakes. It will delve into the foundational concepts of what constitutes a deepfake in both audio and visual domains and explore the intricate technical details of detection methodologies. The tutorial will encompass various aspects of deepfake creation and detection, including the underlying machine-learning models, feature extraction techniques, visual generation pipeline, and evaluation metrics.
A significant focus of the tutorial will be on the interdisciplinary nature of deepfake detection, highlighting the convergence of audio and visual signal processing techniques. Traditionally, the audio and visual domains of deepfake detection have been siloed, leading to distinct research trajectories. By bridging the gap between these two domains, the tutorial aims to develop more robust and comprehensive detection systems by leveraging the synergies between audio and visual techniques.
Participants will also gain practical insights from case studies and real-world examples, showcasing the application of these techniques in detecting deepfakes. The tutorial is intended not only to educate but also to encourage participants to contemplate the broader implications of deepfake technology on society and individual privacy
Day 2 - Auguest 12 (9:45-11:45)
Titile: Defending against Misinformation in the Wild
Abstract: This tutorial provides a concise overview of AI-driven misinformation, covering deepfake images and videos, text-based disinformation, and social media manipulation. We'll explore how these are created and, critically, how to detect them. The session will delve into Natural Language Processing (NLP) for fake news detection and the challenges of social media information manipulation. Finally, we'll discuss integrating cognitive science with deepfake detection to enhance our ability to identify sophisticated AI-generated falsehoods. This tutorial aims to provide a foundational understanding of AI misinformation threats and current detection methods.