Parang Saraf from Virginia Tech Gives a Pecha Kucha Talk on Building Tools for the Future
Discussing a four-year $20 million IARPA funded project that spanned 80 researchers at 13 universities and 3 companies, Parang Saraf details the flagship social media project led by Virginia Tech. The goal of EMBER was to take a plethora of data sources, parse them through a series of machine learning terms and systems to ultimately create asystem predictive of disease outbreaks and protests.
Run as a forecasting tournament, the Virginia Tech-led team was crowned the winner after two years of the project. In order for Saraf’s team to succeed they not only used big data, but also wide data, stemming from many sources including social and online data. Saraf’s team used more unconventional sources such as satellite imagery and restaurant reservations. In the case of the satellite imagery the team examined hospital parking lots to measure the number of cars in the parking lot, while restaurant reservation counts were tracked over time with both sources helping to impact flu trend prediction.
The project was able to successfully predict protests in Brazil, Venezuela, and Mexico 3-5 days in advance, leading them to win the research competition. However, once the project ended, Saraf’s team faced the challenge of taking academic research and using it for commercial applications.The team is exploring themes across different industries.
Potential examples include helping financial industries predict commodity exchange rates and price trends, predicting elections across countries, forecasting cyber attacks utilizing social and online data to make predictions about emerging technology, and within brands to create popularity trend predictions for brand growth. In the last category, the team found nuances across social sentiment’s impact, realizing the predictive quality was limited for brands but highly impactful for entertainment in predicting a TV show’s popularity.
Looking to the future, Saraf and his team are excited to continue growing the applications of their product within commercial industries. Enhancing the predictive power of social data in combination with data sources both big and wide is their ultimate goal.