Experiment various sampling and training methods to build secured deep learning models to strengthen it against adversarial attacks. Develop efficient experimentation methods for parallel training of neural networks maximizing utilization of GPU’s.

Research in the domain of computer vision, optimization of OCR technology, weak supervision and deep active learning. Working on various segmentation models, attribute prediction models, feature extraction, triplet models, disentanglement, metric learning, clustering, image processing, computer vision and snorkel.

Deep learning (DL) is greedy for data and requires a large amount of data supply to optimize massive parameters, so that the model learns how to extract high-quality features. Deep Active learning (AL) attempts to maximize the performance gain of the model by marking the fewest samples. Open source contributor - DISTIL Read More

The first step of creating a virtual hairstyle is to digitize human hair. There is a need to encode human hair information such that each bit in the information tells us something about the hair type or style. For example, we might encode hair information in a 512 bit vector then the first 25 bits might represent the length of hair, next 25 bits might represent color of hair, etc. This kind of disentangled encoding can be used to control individual features of the hair. Read More

Presented at International Conference on Advanced Computing Technologies and Applications. This paper aims on generalizing the effect of number of hidden layers on datasets in Artificial Neural Network. Read More

Building end to end deployable NLP and CV solutions. Working on research problems such as relevant compound extraction, bias removal and technical question answering. Responsible for interaction with subject matter experts to gather domain knowledge and expand use of AI in the respective field

Developed an AI based solution that generates key actionable insights for their clients by automating companies Market and Competitive Intelligence (MICI) services. Designed and implemented user interaction using web stack such as PHP, JS, jQuery for user interaction. Formulated the task of document classification using supervised machine learning for categorizing the documents.

Researched and developed a robust solution for one of their critical projects epub3 automation for global clients to convert their pdf to electronic publication. Formulated an AI solution for detecting different patterns inside a pdf using Luminoth framework. Configured and optimized the object detection model for recognizing patterns inside a book for Scholastic