Deep Learning Library

The CPU only library is derived from this application. The library does not fully implement channels and stride and padding. It doesn’t support recurrent neural networks or LSTMS’s. It is meant for computer vision tasks. It does not support Self-supervised or semi-supervised or reinforcement learning etc. It is an ongoing work. It supports across-layer connectivity. It has been tested for MNIST and cifar10 it may need to be modified for very deep networks. Programming is to be modified for residual network configuration. It implements a network in networks idea by introducing complex units in the convolutional layers. Some ideas like a grid on a convolutional layer also implemented. but ideas around new activation functions have been tried and abandoned. cleanup of code is required. May prove useful to students to design their own custom architectures for computer vision.