Deep Learning Limitations and New Frontiers

Explore the limitations and new frontiers of deep learning in this lecture from MIT's Introduction to Deep Learning course. Delve into the history of AI hype, rethinking generalization, and the capacity of deep neural networks as function approximators. Examine adversarial attacks on neural networks and their limitations. Discover the importance of uncertainty in deep learning and learn about Bayesian deep learning techniques for uncertainty estimation. Investigate model uncertainty applications and the concept of learning to learn. Gain insights into the power of neural networks and emerging ideas in the field of deep learning.

Intro
T-shirts!
Final Class Project
Thursday: Deep Learning in Industry
Friday: Project Presentations
Power of Neural Nets
History of Artificial Intelligence Hype
Rethinking Generalization
Capacity of Deep Neural Networks
Function Approximators
Adversarial Attacks on Neural Networks
Neural Network Limitations...
Why Care About Uncertainty?
Bayesian Deep Learning for Uncertainty
Elementwise Dropout for Uncertainty
Model Uncertainty Application
Motivation
Model Controller
The Child Network
Learning to Learn: A level deeper
This Spawns a Very Powerful Idea