insanity

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

• Based on Theano and NumPy.
• Can be accelerated using CUDA or OpenCL.

##Portable

• Can run on any computing platform:
• Traditional CPUs
• NVIDIA GPUs using CUDA
• Other processing units using OpenCL
• Supports embedded ARM-based systems.