Harmonic Analysis and Machine Learning for Emergency Response

University Of Maryland, College Park, College Park MD

Investigators

Abstract

The research presented in this project resides in threat detection and disaster management. Significant mathematical contributions to this field are still needed, due in part to the ongoing revolution with complex data problems, known as the Big Data paradigm. This project aims to bridge the big data gap, providing added value to the field while simultaneously expanding the impact of modern mathematics. With this point of view, this project studies potentially predictive disaster scenarios, from radioactive leaks, to modern battlefield issues, to natural disasters. However, the notion of threat in this project is not limited to an earthquake, flood, or nuclear explosion, but rather what impedes the people affected by disasters. The project research intends to provide efficient ways of disaster aftermath management. Algorithmic Threat Detection (ATD) is a predictive concept that must be quantified and effectively designed to address major defense problems. After analysis of recent disaster scenarios, a suite of technologies have been formulated that use mathematics to help mitigate disaster impacts. Machine learning and deep learning are major techniques in this suite, and involve data processing, spectral graph analysis, Schroedinger eigenmap technology with customized non-linear potentials, transport models, and recent innovations, ideas, and results dealing with Fourier scattering transforms, pooling operators, and convolutional neural networks. Essentially, this work develops a toolkit from modern applied harmonic analysis and machine learning. The mathematical rigor in this work will enable us to construct fast and efficient disaster mitigation implementations. The ideas and results proposed here are new and innovative, and the applications to threat detection are timely and relevant. These methods may also impact other areas of science.