Projects

Some of my projects done during my time at IITK

Implemented and fine-tuned text summarization algorithms like TextRank, LSA, Google PEGASUS, t5-small
Preprocessed corpora using a variety of NLP techniques like tokenization, stemming, lemmatization
Used ROUGE-N to evaluate accuracy of machine summaries against human summaries for corpus of size 230k
Tested out algorithms and achieved SOTA f1-scores on our datasets for abstractive algorithms after fine-tuning
Created a Flask web application for users to interact with the algorithms and get summaries of articles or blogs

Analysed vaccination rates for covid-19 by race in US counties, along with their association with socioeconomic factors, using a number of statistical techniques
Transformed and pre-processed data by z-score normalization and PCA dimension reduction of feature matrix
Implemented variety of regression algorithms like MLR, Support Vector Regression, Random Forest
Fine tuned the hyper parameters of these algorithms using train and test MSE as metric of measure

Analysed seasonal crop data for last 20 years and unearthed surprising information about the Indian agricultural
Pre-processed 2.5 lakh data points using cleaning techniques to get robust dataset for subsequent steps
Developed new categorical variables of zones and crop to make analysis more understandable and comparative
Used geopandas library to discover the geographical situation of Indian agriculture on zone and crop basis

Modeled a seasonal ARIMA model to forecast next year’s temperatures based on historical data
Checked data stationarity through Dickey-Fuller test, and performed seasonal differencing to ensure stationary TS
Optimized the model parameters using Box Jenkins Method for forecasting purposes
Performed residual analysis and information criterion tests to check model adequacy on the dataset
Yielded absolute MAE of 2.2% on test data for optimized S-ARIMA model

Implemented tweaked version of MH algorithm called Repelling–Attracting Metropolis Algorithm (RAM) for multi-modality
Used Auxiliary Variable approach to derive the steps of the algorithm
Demonstrated how the RAM model outperforms the MH sampler, by generating MCMC samples for real-life numerical examples like sensor network localization, strong lens time delay estimation
RAM provided better simulations for multimodal distributions, as seen by ACF plots, trace plots, acceptance rate and downhill-uphill average proposal numbers of generated samples

Developed expectation-maximization (EM) algorithms to fit multivariate Gaussian mixture models for latent variables
Cross-validated the model on test data to determine optimum values of hyperparameters
Utilized MH algorithm to implement the Markov Chain Monte Carlo (MCMC) method in Bayesian logistic regression model

Conducted structural comparison of monetary authorities in India and the United States
Compared monetary policies of these countries including reserve requirements, discount rate, and open market operations
Critically analyzed various monetary policy strategies employed to control economic variables like unemployment rate, inflation rate, interest rate, GDP, money supply, etc.