Explore the fascinating intersection of neuroscience and artificial intelligence in this lecture by Gabriel Kreiman from Harvard University and Children's Hospital Boston. Delve into the complexities of brain codes and their potential applications in developing more advanced AI systems. Examine the intricacies of visual recognition, pattern completion, and neural mechanisms underlying these processes. Discover how the human brain processes partial information and completes patterns, and learn about the importance of eye movements in scene understanding. Gain insights into neurobiologically inspired models that capture sampling behavior and the potential for creating more robust and intelligent AI systems based on our understanding of brain function.
Overview
Syllabus
Intro
Intelligence: the greatest scientific adventure of all times
Peeking inside the brain
An image is worth a million words
Standing on the shoulders of giants: Mesoscopic connectivity of the primate visual system
Outline
Pattern completion is a hallmark of intelligence
Visual recognition of occluded objects
Visual inference from partial information
Strong robustness to limited visibility
Peeking inside the human brain
Neural mechanisms of pattern completion
Rumination through horizontal connections
Neurobiological recipe: add horizontal recurrent connections to deep convolutional networks
Eye movements are critical for scene understanding
We are legally blind outside the fovea
Four key properties of visual search
Active sampling during visual search
Invariant Visual Search Network (IVSN)
Neurally inspired model captures sampling behavior
Taught by
MITCBMM
