Deepnet

Multipurpose modular deep learning neural network with OpenGl visualization.

Watch the neural net learn. Synaptic weight visualization with OpenGL allows the user to watch
the neural net train.

Dependencies

GLFW 3, OPENGL, GLEW.

Usage

Usage: deepnet /[mode] /[# data] /[size] /[neurons.brain] /[visualize]

mode: /[t/a] either a 't' to specify training mode or 'a' to analyze.
/# data: number of data for training or analysis.
size: the length of the largest data (including solution flag and length header if training).
neurons.brain: file to dump neurons to after training or load neurons for analysis.
visualize: [0 to # data] opengl visualization of training (slower). 0 to disable.

Visualization option

Make targets: spec, line, and tri are available.

spec - Screen clear after each render. Just shows current synapse values.
line - OpenGL lines.
tri - OpenGL triangles.

Primative coordinate system: ( x: synapse0, y: synapse1, z: synapse2 )

Images - Training 100000 iterations

Default after 1.9%

Each colored trail is one layer0 -> layer1 synaptic weight.

SPEC enabled after 3.3%

Each dot is one layer0 -> layer1 synaptic weight.

LINE enabled after 2.1%

Connect multiple synaptic weights with lines.

TRI enabled after 1.7%

Connect multiple synaptic weights with triangles.