I-Corps: Translation potential of enhancing predictions in sparse data environments

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of machine learning technology to bridge crucial data gaps, especially in fields with scarce data. The goal is to focus initially on enhancing yield forecasting for agribusinesses. Currently, addressing critical challenges such as climate volatility, the scarcity of accessible short-term financing, and labor shortages during peak harvest periods is a problem for forecasting. The proposed technology aims to improve business continuity for farmers and streamline operations across various sectors. The commercial applications of this technology may include advanced risk management for agricultural loans, enhanced fraud detection in e-commerce and banking, strategic planning for food policy and security, and the development of sophisticated actuarial tools for insurance companies. Extending beyond the agricultural sector, the technology also has broader applications in energy consumption, finance, and the demand for commodities influenced by external variables such as weather or economic activities. This may improve the agricultural sector's efficiency and foster broader economic impacts through improved decision-making tools. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of the proposed technology. It is based on the previous development of technology in machine learning, combining mathematical simulations with physics-based models to achieve precise time-series predictions in data-scarce environments. Leveraging neural network architectures, the proposed technology is designed to be used in interpreting and enriching sparse data, to provide a tool across various industries. It integrates convolutional neural networks, transformers, and simulation models into a unique predictive framework. This integration enables the algorithms developed to produce highly accurate forecasts with minimal training data, effectively augmenting simulated and real data sets. The ability to process sparse and coarse data efficiently may provide adaptability over existing models, offering significant advancements in fields such as agriculture, insurance, finance, and energy demand forecasting. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.