I-Corps: Faster than Light Big Data Analytics

University Of Texas At Austin, Austin TX

Investigators

Abstract

Data-driven decision making is changing society. Political candidates now collate and analyze databases to identify possible campaign supporters. Couples now meet online through matchmaking websites that pair individuals based on profiles and questionnaires. Marketers now fine-tune their messages based on fine-grain market segmentation provided by extensive web and social network signals. Companies now develop products based on A/B testing or measured customer behavior. However, the diffusion of the basic software platforms and algorithms for analytics has been uneven. Large firms can afford to hire programmers and data scientists to run sophisticated analytics, but most organizations cannot afford the large capital and operational expenses to maintain analytics infrastructure. This leaves them out of the potentially transformative impact of data-driven decision making. This proposal expands the space of high-performance machine learning by harnessing a more general and more efficient parallel programming platform. The large capital expense stems from the lack of a turnkey solution for predictive analytics. Applications are ad-hoc, developed anew based on specific organizational requirements. The large operational expense arises from the data demand of these applications, which requires many machine-hours to produce results. The team has developed a reusable software platform that can parallelize analytics applications with orders of magnitude performance improvements and can run with a fraction of the hardware resources compared to commonly deployed commercial products. On top of this platform, the team has begun to develop a suite of state-of-the-art massively scalable machine learning algorithms. Through the course of this project, the team intends to broadly engage with potential customers of data analytics to help understand possible business models for the proposed technology.