Generation of Images via Attribute Manipulation using Disentangled Representation Learning

Disentangled representation learning studies how a problem can be broken into its constituent components which represent some underlying structure. This aids interpretability of representations learned by deep learning models and allows flexibility in sampling data from generative distributions. In this project, we’ll discuss how the paper on InfoGAN [1] modifies the GAN objective to learn these representations using an information theoretic framework.

Bilateral Filter for Contrast Reduction

The bilateral filter is very versatile with several applications ranging from texture smoothing, to mesh construction to HDR imaging. In this project, we’ll first look at the motivation behind using the bilateral filter for contrast modification and subsequently implement a computationally efficient version for tone-mapping HDR images.

A Primer to Bayes' Classifier

The Bayes’ classifier is a powerful tool in any machine learning practitioner’s arsenal. It’s a generative model and by exploiting Bayes’ rule can be used for classification. In this article we’ll first understand Bayes’ Rule as applied to a supervised classification problem following which we’ll program an algorithm to classify digits using a Naive Bayes’ classifier.