Go to Duke Events
  • Machine Learning Seminar


    Jonathan Su - MIT Lincoln Lab


    Machine Learning, Bass Connections-Information, Society & Culture, Biostatistics and Bioinformatics, Computer Science, Electrical and Computer Engineering (ECE), Mathematics, Pratt School of Engineering, and Statistical Science

    Gross Hall, 330 -- Ahmadieh Family Grand Hall - Map




    Dawn, Ariel





    Under the National Hurricane Program, MIT Lincoln Laboratory is developing HURREVAC eXtended (HVX), which is replacing the legacy evacuation decision-support tool HURREVAC. The platform is designed to enable emergency managers (EMs) make accurate and timely evacuation decisions for their community by providing planning, situation awareness, data analytics, and training tools in a Web browser. HVX provides the novel capability to simulate storm tracks in a Web application with minimal user input or meteorological expertise. This capability will allow EMs to train more frequently and inexpensively, simulate storms that EMs have not already seen, and enable serious games and training analytics to improve EMs' preparedness. The simulator leverages the HURDAT2 database, which contains information on more than 2,800 tropical storms, to generate plausible storms. The intended purpose and the required storm characteristics drove simulator design and technique selection. Storm movement between waypoints is simulated by a Markov model and kernel density estimation (KDE) with a gamma-von Mises product kernel. KDE with a beta kernel generates maximum sustained winds, and linear regression simulates minimum central pressure. Maximum significant wind extents are simulated by Poisson regression and temporal filtering. The un-optimized MATLAB code runs in less than a minute and is integrated into HVX as a Java package.

    Lecture/Talk and Panel/Seminar/Colloquium