Mayur Shastri

As a project under Prof Erik Learned Miller, I generated augmented datasets that had similar distributions to the original datasets by performing Data Augmentation using Deep Convolutional Generative Adversial Networks (DCGANs). Achieved state of the art level Specificity and Sensitivity scores while also leveraging generated augmented images to achieve a top-3 accuracy rate of 78.65% for hemorrhage detection.

Developed a medical application that diagnosed the early onset of Shingles by using a phone camera to perform diagnosis and classification. Built an Inception-ResNet image classification model to detect patterns and outperformed all other classification models with a top-5 error rate of 3.08% on Kaggle. Developed the chatbot interface which focused on analyzing patient symptoms with regards to Shingles and providing valuable first aid.