Though neural networks have been around for a while, I have whittled down the list of players in the game of visualizing them to the following:
- Little understanding of why neural networks work
- We have learned that there is some learning going on at intermediate layers, sort of like intermediate classification because of like objects (example of text)
- Improving performance (Zeiler & Fergus)
By the way these things were achieved
This is the mostly academic work of
- Zeiler and Fergus: Deconvnet
- Jason Yosinski et. al.: DeepVis
- Andrej Karpathy: A Neural Network Playground
...published a paper on visualizing and understanding convolutional networks in 2014, founded in their work deconvolutional networks in 2011
...published a paper and open source tool for visualizing a neural network in real time, with an image or camera feed. This builds on deconvnet.
- Adaptive Deconvolutional Networks for Mid and High Level Feature Learning
- Understanding Neural Networks Through Deep Visualization Paper (Yosinsky)
- ConvnetJS demo: toy 2d classification with 2-layer neural network
- Neural Network Playground
- Understanding Neural Networks Through Deep Visualization (Website)
- These have been produced from viewpoint of the nn engineer or professor, is there an interest for non comp-sci population to visualize NN?
- How to improve interactivity
- Showing how they work
- multiple dimensions
- hook up something like A neual network playground thing with something in real time