The Power and Limits of Deep Learning

Title: The Power and Limits of Deep Learning
Date: Thursday, July 11, 2019 1:00 PM ET / 10:00 AM PT
Duration: 1 hour

SPEAKER: VP & Chief AI Scientist at Facebook and Silver Professor at NYU

Resources:
TechTalk Registration
Deep Learning with Python (Safari book, free book for ACM Members)
Deep Learning Illustrated: A Visual, Interactive Guide to Artificial Intelligence (Safari book, free book for ACM Members)
Grokking Deep Learning (Safari book, free book for ACM Members)
Generative Deep Learning (Safari book, free book for ACM Members)
Deep Learning (Safari book, free book for ACM Members)
Practical Deep Learning for Cloud and Mobile (Safari book, free book for ACM Members)
Deep Reinforcement Learning and GANs: Advanced Topics in Deep Learning (Safari video, free video for ACM Members)

Question on the Power and Limits of Deep Learning:

Dear LeCun,

The state of the art in deep learning has been inspired by the way human brains work. Sometimes, however, the analogy between artificial and natural intelligence seems to fall apart. The first example that comes to mind is how convolutional neural networks are based on how humans recognize objects in the visual cortex. However, CNNs are not exact models of the human visual system. Given these kinds of scenarios, do you still see the paradigm that connects human brain function to ANNs as a promising one for deep learning going forward? Or does this suggest that we know less about human brain function than we thought?

What do you think about the feasibility and scope of model compression and pruning in cnn for inference on resource constrained mobile devices?