What's conventionally described as the trellis structure of the connections between two fully-connected layers of a neural network is really a visual description of a matrix-vector product. Driven by the belief that mathematics are beautiful when properly represented, I'm building a little tool to quickly generate SVG representations of these trellises, to visualize classic fully-connected neural networks, on top of d3.
In the future I'd like to have three layouts:
- Conventional FCNN layout
- LeNet CNN layout
- AlexNet layout (via webGL)
