Generative AI, Computer Vision, Apple Vision Pro

Supervised and Semi-supervised learning techniques for Alexa NLU.

Interning at Machine Learning platform team for Apple

Performed domain adaptation for text and video time-series data using Convolutional Neural Networks, Long Short Term Memory networks and Adversarial Learning. Built an end-to-end steering system on simulation of driving video and learnt steering angle prediction on real-world driving videos.

As a deep learning intern at XRCE, I worked on domain adaptation for time-series data like text & videos. I am using LSTMs and Adversarial Learning to train a domain-invariant deep learning model. I built a sequence tagger for Wall Street Journal data and automatically learnt a sequence tagger for medical data.

Built a state-of-the-art sequence-to-sequence tagger for citation strings in scientific documents using graphical models like Conditional Random Fields and deep learning models like Long Short Term Memory. Used word and character embeddings which eliminate the requirement of hand-designedfeatures and makes the learning minimally supervised. Created a docker image of the system to make the tagger environment independent. Research paper for the project published in International Journal on Digital Libraries 2018.

Developed an end-to-end discourse parser which identifies the relation between two discourse units (like sentences), extracts the phrases from the sentences and classifies the semantic sense of relation between these phrases. Participated and published results of the system in Conference on Natural Language Learning, 2016. Presented the results of the parser at ACL, 2016.

Extended PyWikiBot support to wiki websites which fall under WikiMedia umbrella. Made the process of PyWikiBot deployment automatic as compared to it's previous procedure of using static, hand-written configuration files.