The simple, minimal deep learning library for Python, intended to be an alternative to libraries such as Caffe and Torch7.
Insanity is licensed under Creative Commons Zero.
##Powerful
- Provides an API for neural networks with several layer types:
- Fully-connected
- Convolutional
- Max-pooling
- Provides several neuron activation functions:
- Linear
- Sigmoid
- Tanh
- Rectified linear unit
- Implements stochastic gradient descent for training neural networks.
- Provides a dropout function for increasing the reliability of network layers.
- Provides methods for serializing and un-serializing networks using JSON.
##Fast
##Portable
- Can run on any computing platform:
- Traditional CPUs
- NVIDIA GPUs using CUDA
- Other processing units using OpenCL
- Supports embedded ARM-based systems.