Project 2: Dewang Sultania

[Iron-Stark]

• Repo Link

• Features:

  • Implemented CPU, Naive, Thrust and Work Efficient scan
  • Implemented Forward propagation and backward propagation for Multilayer Perceptron. Got all 52/52 correct for the given dataset.
  • Apart from the data given, trained the network on MNIST dataset of 60,000 handwritten digits and achieved 95.65% accuracy on that.

• Feedback: The MLP assignment could have been better structured but I had fun implementing backpropagation from scratch. The one thing that I would like to recommend for future is having more than 1 examples per class for training as training on the same dataset as given is not actually learning but memorization. What I had fun with the most is training it on MNIST. I guess that can be given as the assignment in future iterations of the course.