
Abstract: This talk will discuss common signal processing techniques on time series data that is to be used for machine learning applications. In most applications, choosing the best representation of data, either through transformation or extraction of relevant features, can be a very effective way to improve a machine learning model by vastly reducing the size of the model’s input data set, removing extraneous information, or choosing an alternate basis. Signal processing is a set of tools developed for time series data which aims both to remove imperfections and to extract relevant information without distorting underlying trends. It has far-reaching applications across a wide range of disciplines, from the physical sciences (acoustics, electronics, navigation) to finance, and is often required in machine learning pipelines designed to make conclusions based upon time series data.
In this talk, we will discuss a number of common tools in signal processing and will show how many of these tools can be implemented in various Python packages. At the heart of our discussion will be an introduction to various types of filtering, tools that are used to remove degradation or “noise” from time series data to find an underlying trend. While the tools that we will discuss can be applied more broadly, we will primarily draw upon examples from our work in mobile telematics, including applications such as data cleansing for inertial sensors, bias removal, and sensor fusion between inertial sensors and global positioning systems. Demonstrations will be run in Jupyter to show how to quickly make use of available tools in SciPy and other open source packages."
Bio: Keith Santarelli is a data scientist at Agero where he focuses on applying sensor fusion to making our nation's roadways safer. He holds a Ph. D. in Control and Systems Theory, and an S.B. and M.Eng. in Electrical Engineering and Computer Science, all from the Massachusetts Institute of Technology. Prior to joining Agero in 2016, he spent a decade working in the government defense and energy sector as a signal processing engineer, working on mission-critical problems such as sniper detection, border safety, robust navigation in hostile environments, and anti-jamming technologies for satellite navigation systems.
Dr. Santarelli focuses on the application of time series analysis techniques for both engineering design (e.g., designing algorithms for mobile devices) and business analysis. An avid applied mathematician, he is passionate about putting mathematics into practice for the benefit of all, and he enjoys speaking with aspiring scientists and engineers about the possibilities for their own careers in data-centric roles.