04 Sep 2015
By Sam Green
Supervised by Professor William Knottenbelt and Dr. Jack Kelly
This thesis contained in this repository explores visualisation as a method to shed light upon how deep neural networks during training in order to assist researchers more effectively train and debug their models.
The thesis is divided into two parts.
- A survey of existing literature into neural networks and visualisation techniques.
- An exploration of a novel approach to visualising deep neural networks using a combination of t-SNE and PCA to reduce the dimensionality of the data and D3.js to visualise and make interactive the results.