I'm a PhD student at Johns Hopkins and a graduate of Georgia Tech. I'm interested in using natural language processing to make useful software. I've worked on projects involving machine translation, document level NLP, data-to-text generation, and model interpretability. Before coming to Johns Hopkins, I worked with Microsoft's Machine Translation research group.
My research interests broadly include translation and multilingual language understanding. Recently, I've become interested in efficient neural models, cross-lingual information extraction, and domain adaptation.
What I Do
What I've Done
Recent jobs, internships, and campus involvement during my undergrad.
Semantic Machines is a startup acquired by Microsoft in 2018. This group works on executable semantic parsing and conversational AI for task oriented dialogue.
I worked with the Machine Translation research group at Microsoft before starting my PhD. I worked on research projects involving production quality neural models, semi-supervised learning, and data to text generation. [Poster] [Paper]
I also worked on low resource domain adaptation recipes for production systems.
I worked in the Computational Linguistics Lab at Georgia Tech under Dr. Jacob Eisenstein. I previously worked on document level tasks, like translation and summarization. Prior to graduating, I worked on a project analyzing learned representations in character level tagger models. This work was presented at BlackboxNLP 2019. [Code] [Paper]
I worked on the Bing team at Microsoft. I built a real time performance analysis system to assist in diagnosing anomaly root causes for services.
I assisted students in understanding AI topics including heurisitic search, probablistic reasoning, and neural networks.
The Agency is the AI/ML reading group and club at Georgia Tech. I gave weekly talks on machine learning and organized club events and workshops. The ML center wrote an article about us!
I worked at Quantlab Financial, an algorithmic trading firm. I used signal processing and unsupervised learning methods to detect and correlate performance anomalies across the trading systems.
I was a TA for CS 1331, an introductory Java course. I planned and taught recitation lectures to 50 students weekly.
Projects I've done for courses and fun, primarily in undergrad. Code for some is available on Github
As a member of the organizing team for HackGT, I built a photo booth that uses neural style transfer to transform a picture. Participants at our events can use the physical booth or text in photos to our MMS service.
Airbnb vs. Hotels
This was a course project for Data and Visual Analytics. Our project was an interactive visualization showing the differences between Airbnb listings and hotels in New York. I built a directed sentiment analysis system that tried to quantify sentiment towards specific aspects of a listing from reviews, like hospitality or cleanliness.
Paizza: Winner @ Hack Duke 2016
Paizza is an improved search engine for Piazza, a commonly used course forum platform. I built an information retrieval and document clustering backend system for finding similar questions on a forum. Our project won first prize in the Education Track!
DM me on Twitter, or email me at mmarone [the number one] at jhu dot edu