Multi-Relational Graph Neural Network for Darkweb Actor Classification
Multi-Relational Graph Neural Network for Darkweb Actor Classification

I propose a semi-supervised deep learning framework to detect key actors by classifying underground forum users into different groups based on their activity patterns.

TensorflowKerasGCN
Quark Gluon Classification using Contrastive Learning Representations
Quark Gluon Classification using Contrastive Learning Representations

I trained a transformer-encoder network with contrastive loss to map particle jets to a representation space. I implemented ML models to classify the obtained representations of jets into quarks and gluons.

ScikitXGBoostPytorch
Graph Neural Network for the XENON Detector
Graph Neural Network for the XENON Detector

I designed a graph neural network to reconstruct the position of collisions happening in the XENON detector, an underground experiment for studying potential dark matter candidates.

Graph Neural NetworkRegressionSpektral
Super Resolution CNN for Sea Surface Temperature
Super Resolution CNN for Sea Surface Temperature

High-resolution measurements of sea surface temperature (SST) are crucial to climate and ocean state forecasting. However, on cloudy days, satellites can only procure low-resolution SST measurements. I implemented a super-resolution Convolutional Neural Network to successfully downsample Sea Surface Temperature (SST) fields of the Bay of Bengal from a spatial resolution of 15 km to 1km.

CNNComputer VisionTensorflow
Automatic Realtime Machine Translation
Automatic Realtime Machine Translation

I designed a machine translation model capable of translating short sentences from five languages (Chinese, Persian, Sinhalese, Burmense and Indonesian) to English and vice versa.

GRUNatural Language ProcessingTensorflow
PyTorch based Model for Text Recognition and Detection
PyTorch based Model for Text Recognition and Detection

Fancy fonts in movie posters attract humans no doubt, but pose an equally huge challenge to text detection and recognition models. For this, I integrated a scene text detection network with a text recognition network to create a single pipeline that could read "scene-text" off images.

OCRComputer VisionPyTorch
Stock Price Prediction
Stock Price Prediction

I experimented with different techniques (inclusive of Facebook Prophet, Naive Forecasting, Neural Networks and XGBoost) on a single dataset to learn the ins-and-outs of time-series forecasting.

Time Series ForecastingXGBoost
Simulation of Galactic Evolution

I researched how galactic evolutions are carried out. The end goal of the simulation was to draw a comparison between observed and theoretical isotopic mass ratios. Comparative analysis of the stellar nucleosynthetic yields, infall rate formulations, etc., adopted by existing galactic simulation approaches was performed. I was supervised by Dr. Kuljeet Kaur Marhas, Physical Research Laboratory.

Python
Image Classification on CIFAR-10
Image Classification on CIFAR-10

Classic beginner-level project that every deep learning enthusiast begins with. I compared the performance of a vanilla Convolutional Neural Network with that of a pretrained ResNet Model on different CIFAR-10 image sizes. This project exposed me to the intricacies involved in training a CNN.

Computer VisionCNNClassificationTensorflow