VISUALISATION (2016)

In VISUALISATION, I try to create a virtual 3D sculpture to model the firing of virtual neurons inside a virtual brain. The inspiration for this was the workings of the Character Recursive Neural Network. This system was somewhat easier to understand than other neural networks I have previously experimented with as it's input and output are purely textual, and therefore easier to understand. The program effectively reads the input text repeatedly, building up links between characters, or strings of characters, in much the same way as the human brain builds up links between sensory inputs in the form of synapses and neurons.

When I started building the program, I intended it to be 2D, as it was to be projected onto a screen. As it evolved however, I realised the jump to 3D was smaller than I had imagined, so I took it. I was very pleased with the result, although the more complex the structure becomes, the larger the processing burden on the computer. Because of this I had to invent processor saving tricks to achieve a similar goal with fewer draw calls. The main bottleneck was the the line drawing part of the program as it had to draw a line from every vertex, to every other vertex, every frame. I decided to combat this by allowing the lines drawn to be chosen at random, but for only 8000 or so to be drawn so that the amount of draw calls was limited. This had the effect of making the lines flicker. it also lead to an interesting phenomenon in which overlapping lines would amplify each other. The Lines which carried more connections appeared brighter than those with fewer connections.