Watching Neural Networks Learn: Peeking Under the Hood During and After Training


Tali Beynon (Lead Developer, Deep Learning at Wolfram) is the developer of the high-level neural net framework that can be found in recent versions of Mathematica.

In this presentation I'll try to dispel the myth that neural nets are black boxes, and show some examples of how we can 'peek under the hood' of a net during and after training (using Mathematica). After seeing these visualizations, I hope you'll leave with a better intuition for the overall training process, the behavior of specific kinds of nets, and how they succeed (or fail) to learn.

Zoona has kindly volunteered that we can host this meet-up at their office - 1830 for 1900.