Implementation of diffusion models with varying noise distributions (Gaussian, GMM, Gamma) and scheduling techniques (cosine, sigmoid) to assess generative performance using KL divergence and dynamic scheduling approaches.

Using Monte-Carlo simulated datasets, a completely transparent Boltzmann Machine trained on 1-D Ising chain data is implemented to predict model couplers in the absence of past coupler values. Methods from machine learning applied to theoretical physics are on display in this work.

Implementation of a Denoising Diffusion Probabilistic Model with some mathematical background.

Kullback-Leibler divergence in Python

My lab work of “Generative AI with Large Language Models” course offered by DeepLearning.AI and Amazon Web Services on coursera.

Implementation of the Non-negative Multiple Matrix Factorization (NMMF) algorithm proposed in Takeuchi et al, 2013 with some modifications. There is a python native version NMMFlexPy and a R wrapper NMMFlexR

Text Analytic Projects

average-KL-divergence-calculator.py is a Python script that calculates the average KL divergence for each FASTA file in a directory and produces separate output files and a combined output file with the results.

Unlocking the Power of Generative AI: In-Context Learning, Instruction Fine-Tuning and Reinforcement Learning Fine-Tuning.

Forward Sampling-Conversion of BN

Generating gray scale images of numerical digit by using variational autoencoder

The Unstable Population Indicator

Scheduling TRPO's KL Divergence Constraint

Novel technique to fit a target distribution with a class of distributions using SVI (via NumPyro). Unlike standard SVI, our "data" is a distribution rather than a finite collection of samples.

Variational autoencoders using Kera's modular design

Factorized variational approximation using a univariate Gaussian distribution over a single variable x.

Some code to Get the Optimal relative Transport started. This will be slowly updated if needed.

Python information theory computation

The Dirichlet Mechanism for Differentially Private KL Divergence Minimization

In this project, we explore how we can use entropy and information in language models and how we can optimize it for generative tasks.