Comparative study of learning approaches

Compared supervised, semi-supervised and self-supervised approaches on image classification task on STL 10 dataset. Used a ResNet-18 model for supervised, vanilla CNN for semi-supervised and SimCLR for unsupervised learning.

Meta learning library using JAX

This project aims to implement various Meta Learning methods and benchmark them on few-shot learning tasks. The goal is to build a modular extensible and well documented library of these methods in JAX.

Probing language inference models

Trained a classifier model on SNLI Dataset to determine the inference relationship between the sentence pairs and performed probing on thee model’s layers for POS tags.

Source code synthesis

Developing a model to perform code search, syntax error detection and code repair on different programming languages.