ashenoy95 / ml

Machine Learning, Neural Nets, Deep Learning basics

Machine Learning

• speech-denoising: MLP and RNN based speech denoiser implementations using TensorFlow.

• em.py: EM algorithm for a Mixture of Gaussians on one-dimensional data.

• mnist-shallow.py: TensorFlow implementation of MLP (1024x5) using ReLU activation, He initialization and Adam optimization giving over 98% accuracy. Also analysing TSNE & PCA.

• nw_compression.ipynb: Network compression of 'mnist-shallow' using low rank approximation of trained weights (using the top 20 singular values after SVD), thereby using only about 4% of the memory of the original network.

• parity.py: MLP (4-4-1) with backprop from scratch to solve the parity problem for a 4-bit input.