Pranav Nair | projects

Leveraging Commonsense Explanations for Question Answering

T5 trained on ANLG and COS-E datasets to generate explanations. The model obtained from the previous training stage is used to generate explanations for ReColr contexts. The explanations, along with the context and question, passed to ALBERT for prediction.

Experiments on CoLA

A Transformer and BiLSTM architecture are both coupled with ELMo to evaluate performance on the CoLA (Corpus of Linguistic Acceptability) dataset.

SCRF for Name Entity Recognition

A semi Markov Conditional Random Field for Named Entity Recongnition.

A Seq2Seq Model for Generating Explanations

A simple LSTM and dot product attention-based model for generating commonsense explanations. Trained on COS-E to generate explanations.

DCN+

Implenetation of Dynamic Coattention Networks + Mixed Reinforcement learning objective(self critic) and question answering.

BiDAF

Implementation of Birectional Attention Flow in PyTorch for SQuAD question answering.

QRNN

Qusai Recurrent Neural Network for Sentiment Analysis on IMDB datset.

GA Reader

Implenetation of Gated Attention Reader for Cloze style question answering and reading comphrehension.

DeFINE AWD LSTM

Implementation of DeFINE word embeddings with AWD-LSTM for language modeling. The input and output embeddings for AWD-LSTMM are tied.

HeartCare

Heartcare is a diagonostic management project developed in Django. Admin can add doctor and add services. User can view doctor profiles and they can also book appointments.

Housing Complex Management System

A Housing complex managament system developed with Spring MVC.