Neha Patki
MIT EECS - Accenture Undergraduate Research and Innovation Scholar
Machine Learning to Solve Word Problems in Finance
2014–2015
Regina A. Barzilay
This project aims to further research in automatic problem solving techniques. Currently, in ‘Learning to Automatically Solve Algebra Word Problems’, Kushman, Artzi, Zettlemoyer, and Barzilay describe an approach for solving basic algebra word problems by extracting variables and probabilistically determining a template of equations. Using similar techniques with natural language processing and machine learning, this project will extend the current model to solve finance-related word problems. This will involve creating features to express the new knowledge needed when working in the field. Exploring the finance domain will help us better compare structure, and recognize everyday language patterns for encoding the mathematical information needed to solve a problem
This past summer, I worked on machine learning for scraper detection at Yelp, which analyzed traffic patterns to perform probabilistic analysis and classify incoming search requests. I am also minoring in business and I am interested in the mathematics of finance. For a side project, I’ve worked on a service that ingests daily stock data to automate an optimization for asset allocation