Beyond Big Data: Real-Time and Low-Storage Dynamical Identification for Big Data Applications

Web Published:

Princeton Docket # 14-3043

Researchers in the Departments of Mechanical and Aerospace Engineering and Applied and Computational Mathematics at Princeton University have developed a general framework that enables efficient trend analysis for large and streaming datasets.  The innovation extracts dynamical features from available data, thus reducing the dataset to a set of structures sorted by temporal characteristics.  These structures correspond to fundamental signatures embedded in the data that can be classified by growth/decay rates and oscillation frequencies.  Access to this set of dynamical features simplifies the task of identifying growth/decay/oscillation rates in subsets of the data, while providing a systematic measure for event classification.  The extracted temporal structures also provide a means of predicting future data.


The “low-storage” nature of the method makes it a viable tool for analyzing and forecasting large amounts of data, such as those encountered in applications of market research, online dating, and internet ad placement.  The extracted dynamical features provide a means of determining, for example, where an internet ad campaign has had and will continue to have most influence.  Moreover, the streaming nature of the framework, along with its ability to handle noisy measurements, makes the method well-suited for use in applications that require real-time processing of potentially noisy measurements, such as bio-medical devices, video processing software, fault detection and structural health monitoring systems, building energy control systems, aircraft wake hazard detection technologies, fluid flow diagnostic tools, laser characterization utilities, stock market day trading software, and more.



     Medical devices

     Video processing

     Real-time adaptive control

     Real-time flow diagnostics

     Internet ad placement

     Online dating


       Real-time forecasting/trend analysis from large datasets.

        Dynamical feature extraction for analysis and classification.

        Predict future trends from available data.

        Refine predictions as new data become available.


Maziar S. Hemati, Postdoctoral Research Associate, Princeton University

Dr. Hemati is currently a Post-Doctoral Research Associate in the Mechanical and Aerospace Engineering Department at Princeton University, where he is developing algorithms for efficient modeling and control of complex systems, such as wind farms, neural networks, and biologically-inspired robotic swimmers.  He earned his PhD in Mechanical Engineering at UCLA in 2013.  His doctoral research focused on developing novel techniques for aerodynamic modeling and sensing, including strategies for wake detection and localization in aircraft formation flight missions and techniques for predicting aerodynamic forces in response to aggressive flight maneuvers.

Matthew O. Williams, NSF Postdoctoral Fellow, Princeton University

Dr. Williams received his PhD in Applied Mathematics from the University of Washington in 2012.  While at UW, he worked with Professor J. Nathan Kutz to develop data-driven procedures for model reduction in mode-locked lasers and other systems in nonlinear optics.  At Princeton University, he has continued to focus on methods that use data to understand dynamics.  In collaboration with Professors Clarence W. Rowley and Ioannis G. Kevrekidis, he has been developing algorithms that combine ideas from machine learning with mathematical concepts from dynamical systems with the goal of understanding complex systems without the need for explicit equations.

Clarence Rowley, Professor, Department of Mechanical and Aerospace Engineering, Princeton University

Professor Rowley is a recognized leader in dynamical systems analysis of large-scale systems, and his awards include a CAREER award from the National Science Foundation and a Young Investigator award from the Air Force Office of Scientific Research.  In particular, his research has led to advances in techniques for modeling and controlling fluid flows in a variety of contexts.



Intellectual Property and Technology Status

Patent protection is pending.

Princeton is seeking to identify appropriate partners for the further development and commercialization of this technology.  



Michael Tyerech

Princeton University Office of Technology Licensing • (609) 258-6762•

Laurie Bagley

Princeton University Office of Technology Licensing • (609) 258-6762•





Patent Information:
For Information, Contact:
Michael Tyerech
former Princeton Sr. Licensing Associate
Princeton University
Maziar Hemati
Matthew Williams
Clarence Rowley
big data
medical device