interactive-neurons

Author: Simon Liu - simonl2@illinois.edu

interactive-neurons is an interactive GUI that allows you to design and train simple convolutional neural networks using the MNIST dataset. Through the GUI, you can tweak the network architecture, including features such as layers, connections, and activation functions. You can also customize training parameters such as the learning rate, training time, and optimizers.

Build Requirements:

1. CMake
2. Cinder 0.9.3dev
3. imnodes CinderBlock
4. flashlight

Build Instructions:

1. Clone Cinder 0.9.3dev with instructions from https://github.com/cinder/Cinder. Refer to this directory as $CINDER.
2. Clone imnodes CinderBlock repository into $CINDER/blocks.
3. Install flashlight library and any requirements.
4. Create a subdirectory $CINDER/projects. Clone this repository into the subdirectory.
5. Build Cinder using CMake.
6. Build this project using CMake.
7. Download and extract MNIST data files into $CINDER/projects/final-project-imonlius/assets/mnist.

Usage:

1. Build and run cinder-interactive-neurons target to run interactive-neurons.
2. Use Network Editor window to design model architecture. Right click to open menu to add layer nodes and use mouse to drag links between nodes.
3. Use Train Model under Menu to begin model training configuration. Training log and/or exceptions will appear in the Log window.