Audio-Visual Self-Supervised Learning: Insights from Infant Development

Explore the fascinating world of audio-visual self-supervised learning in this 48-minute talk by Andrew Zisserman from Oxford University. Delve into how recent advancements in computer vision address the challenge of multimodal learning, as highlighted in the classic paper "The Development of Embodied Cognition: Six Lessons from Babies." Discover how neural networks can learn to synchronize audio and visual streams from video data without manual supervision, mimicking an infant's learning process. Gain insights into the network's ability to localize speaking faces and sounding objects, demonstrating the power of this approach in understanding lower-level intelligence from AI, psychology, and neuroscience perspectives.