PROJECT 1: BERT Based Classifier for Causal reasoning Trained a classifier model using the BERT transformer that would predict whether the given pair of text and reason are related or not. Performed basic EDA, data augmentation using negative sampling (used cosine similarity), training and reported metrics like f1 score, recall, support and precision. Improved efficiency using hyperparamater tuning (grid search).
PROJECT 2: FEDERATED LEARNING MODEL Trained a classifier model on the federated learning setting using the Flower Framework implemented on the MNIST dataset on 108 clients. Optimised the model by implementing various aggregation strategies like FedAvg, FedOpt, SCAFFOLD etc.