Explore a 52-minute lecture on feedforward and feedback processes in visual recognition presented by Thomas Serre from Brown University's Cognitive, Linguistic & Psychological Sciences Department and Carney Institute for Brain Science. Delve into the limitations of convolutional neural networks in visual reasoning tasks and discover a novel recurrent network model inspired by the visual cortex. Learn how this computational neuroscience model addresses shortcomings in state-of-the-art feedforward networks for complex visual reasoning. Examine topics such as computer vision achievements, adversarial attacks, ImageNet, computational neuroscience, and the potential contributions of neuroscience to artificial intelligence. Gain insights into the depth of processing, experimental data, and the benefits of this approach through discussions on semantics, cluttered ABC results, and proof of concept.
Feedforward and Feedback Processes in Visual Recognition
Overview
Syllabus
Introduction
Computer vision achievements
Adversarial attacks
Our own visual system
Deep Neural Network
ImageNet
Shattered ImageNet
Training accuracy
Depth of processing
Computational neuroscience
Three key ingredients
Experimental data
Whats the point
The benefit
Semantics
Cluttered ABC
Results
Proof of concept
Conclusion
Taught by
MITCBMM
