Second Assignment of E1-313 (TIPR)

Deep Neural Network using Numpy

Present Functionality

  1. Evaluated on Popular Datasets like MNIST, Cat-Dog and other datasets
  2. Entire implementation done from scratch using NumPy only
  3. Performance & Quality of models compared with Keras using Tensorflow

Features to be added

  1. Some memory optimization can be done to work in 1/3rd Memory
  2. Dynamic loading of datasets will be enabled for handling large datasets
  3. Numerical stability in Sigmoid and other functions
  4. Variety of Cost functions can be added
  5. Adding Batch normalization and Dropout
  6. Warm training
  7. GPU accelerated Numpy support on CUDA & cuDNN