ODSC East 2016 - Building Neural Networks at Twitter with Autograd

Explore the cutting-edge world of neural networks and automatic differentiation in this ODSC East 2016 conference talk by Alex Wiltschko, a neuroscientist-turned-engineer at Twitter Cortex. Dive into the challenges of gradient-based models and learn how autograd, a machine learning-oriented implementation of automatic differentiation, revolutionizes the process of building and training neural networks. Discover how this powerful tool simplifies the creation of even the most exotic neural architectures, making it accessible and efficient for model builders. Gain insights into the state of machine learning software, computational graphs, and the future directions of AI development. Perfect for those interested in advancing their understanding of deep learning frameworks and seeking innovative solutions for gradient derivation in complex models.

Introduction
Outline
State of software
Numerical computation
Machine learning
Abstraction quality
Restricted Neural Networks
Neural Network Internals
Neural Network Training
Computational Graphs
Modularity
Most granular approach
Autograd API
Neural Network Models
Extreme Flexibility
Features
Future directions
Machine learning textbooks