Shahrukh Quraishi

The objective of this project is to assess the performance of linear regression, its regularized variant, and the Bayesian model, incorporating model selection.

This algorithm is implemented from Bishop: [https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-and-Machine-Learning-2006.pdf]

This project draws inspiration from Conditional Variational Autoencoders. Primarily, I have incorporated two main concepts: the first involves Professor Roni, an IU professor, and the second entails a modification of a semi-supervised learning approach using a deep generative model as outlined in the Google deepmind research paper found at https://arxiv.org/pdf/1406.5298.pdf.

This report showcases the implementation of the aforementioned methods and provides a comparative analysis of… Show more

This project draws inspiration from Conditional Variational Autoencoders. Primarily, I have incorporated two main concepts: the first involves Professor Roni, an IU professor, and the second entails a modification of a semi-supervised learning approach using a deep generative model as outlined in the Google deepmind research paper found at https://arxiv.org/pdf/1406.5298.pdf.

This report showcases the implementation of the aforementioned methods and provides a comparative analysis of their results against those presented in the referenced papers. Show less

In this comprehensive exercise, I apply both forward and backward normalizing flows to assess their bijection properties. Additionally, I conduct density estimation on a dataset, exploring how the estimation is influenced by varying parameters.

This exercise is guided by the principles outlined in Kevin P. Murphy's book chapter on normalizing flows.

Link: https://probml.github.io/pml-book/

This project adopts a variational probabilistic algorithm, grounded in the formulation outlined in the paper found at https://www.stats.ox.ac.uk/~teh/research/bayesml/kddcup2007.pdf. The algorithm builds upon the traditional SVD matrix factorization approach by introducing priors on both the U and V matrices. The estimation of the posterior distribution is achieved through the utilization of Evidence Lower Bound (ELBO) and the variational Expectation-Maximization (EM) algorithm.

I… Show more

This project adopts a variational probabilistic algorithm, grounded in the formulation outlined in the paper found at https://www.stats.ox.ac.uk/~teh/research/bayesml/kddcup2007.pdf. The algorithm builds upon the traditional SVD matrix factorization approach by introducing priors on both the U and V matrices. The estimation of the posterior distribution is achieved through the utilization of Evidence Lower Bound (ELBO) and the variational Expectation-Maximization (EM) algorithm.

I present the implementation of this algorithm on the MovieLens dataset, conducting experiments with various hyperparameters to assess the model's performance. Show less