Pranay Seshadri
Georgia Institute of Technology
Associate Professor in Aviation & Machine Learning,
Daniel Guggenheim School of Aerospace Engineering,
Faculty, Center for Machine Learning
Georgia Institute of Technology.
Email | Publications
Teaching
Spring 2024
AE2010/2011: Fluid Fundamentals & Thermodynamics
AE8803: Gaussian Processes for Machine Learning
Research
I am broadly interested in how machine intelligence can identify pathways to net-zero aviation. Surprisingly, if you ask certain large language models this question, you get a fairly concrete response.
My work involves building probabilistic models to understand what is currently happening within an ecosystem (engine, airplane, airport) and to forecast what is probably going to happen. This is particularly important in trying to find solutions to sustainability challenges. Some of my ongoing projects include:
Probabilistic pattern of life modeling in an airport
The goal of this project is to use years of historical data to infer the movement of various assets on the apron, both today and in the future. Such a model of life can offer ways to more accurately track apron vehicle emissions, predict airport energy requirements, check violations of safety regulations, and even evaluate new policy scenarios. The advantage of such problems is there is no lack of past data; the only difficulty is in identifying how best to synthesize troves of data.
Keywords:gaussian processes, instance segmentation, CCTV, ADS-B
Assessing the environmental impact of future flights
There are numerous carbon (and other emission) calculators based on great circle trajectories. I am interested in how we can combine past real trajectories with weather data (including particulates) to better predict emissions for future trajectories. Here too, there is no shortage of certain types of data. Models built using this kind (and scale) of data can give us a better understanding of the environmental impact of future flight routes.
Keywords:gaussian processes, ADS-B, weather models
Enhancing measurements with machine intelligence
I am interested in how we can utilize the predictive capabilities of different machine learning tools to better calibrate and transfer information between sensors. Aside from filling in the blanks between typically sparse spatial sensor data, given relevant sensor time series histories, is it possible to forecast what a sensor will measure?
Keywords:gaussian processes, Kalman filtering, deep autoregressive models
Contact
Laboratory
Room 108 Montgomery Knight
270 Ferst Drive,
Atlanta, GA 30332