Project

Natural Language Inference

Solved NLI problem using logistic regression and DL based approach to compare

SRGAN

Implemented SRGAN for upsampling images. Tried different architecture and training processes to improve performance

Speaker Recognition

Worked on GMM-UBM and advanced I-vector based approaches for speaker recognition

CNN Pruning

Implemented filter prunning for CNN and proposed similarity based approach to further prune model without affecting performance

Improving sample complexity of HRL algorithms

Worked with team to make learning of higher level policy faster by making transitions more denser by considering sub-transitions.

Word Recognition

Considered phoneme level Hidden Markov Model (HMM) with continuous gaussian distribution for word recognition