Bayesian Neural Network for the Noisy XOR problem

This project was done as a submission to the SAiDL induction assignment.

Code

Abstract: Bayesian ML is a branch of ML relating to creating statistical models based on the Bayes Theorem. Through this project, I learnt about Bayesian ML and Metropolis-Hastings Sampling Algorithm. I created a 2 layered Bayesian NN with warm starting modeled around a 2 layered MLP for the XOR problem. I was able to get a best accuracy of 84.4% on the testing data when I trained on 10,000 data points for 2000 iterations. This was better than the base accuracy of 55% obtained through random choice of the weights.