For my PhD research at the Robotics Institute at CMU from 2010-2015, I worked with the CMU Auton Lab on the Academic Research Initiative (ARI) project developing the Bayesian Aggregation (BA) data processing framework.
Bayesian Aggregation (BA) is a method for characterizing variability in noisy sensor data, allowing robots to better perceive the world in terms of detecting sources and signals of interest. The machine learning framework allows extracting maximal information from single sensor observations and then fusing multiple observations together to detect and characterize the properties of a signal generating source or process of interest.
The research is illustrated with application to the Nuclear Threat Detection domain, allowing Homeland Security and the Domestic Nuclear Detection Office (DNDO) to analyze the large amounts of spectrometry data that can collected by mobile sensors in real time to keep the U.S. safe from threats. We are also working to generalize it to many other domains. Check out some of our contributions below.
I successfully defended my PhD thesis on July 13, 2015! You can download my PhD thesis below:
You can also view my earlier thesis proposal (November 2014) and quals (March 2013) materials here:
Here are my journal papers related to Bayesian Aggregation research:
Tandon, Prateek, Peter Huggins, Rob Maclachlan, Artur Dubrawski, Karl Nelson, and Simon Labov. Detection of Radioactive Sources in Urban Scenes Using Bayesian Aggregation of Evidence from Mobile Spectrometers. Elsevier Special Issue on Mining Urban Data. Information Systems. 2015.
Here are my publicly available conference presentations:
The work has produced no fewer than 17 conference posters in 5 years! Is that a PhD record? Here are most of the poster presentations (broken up by topic area):
Presenting Bayesian Aggregation (BA) and Poisson Modeling capabilities to neural big data and experimental physics communities for generalization of method:
Tandon, Prateek, Peter Huggins, Artur Dubrawski, Simon Labov, and Karl Nelson. Poisson and Bayesian Estimation of Low Photon Count Signal Source and Noise Components. BigNeuro Workshop at Neural Information Processing Systems Conference 2015.
Tandon, Prateek, Peter Huggins, Artur Dubrawski, Simon Labov, and Karl Nelson. Building a Robust Detector Algorithm: Application of Bayesian, Nonparametric, and Poisson Methods to Improve Photon Denoising. Aleph Workshop on Machine Learning for Experimental Physics at Neural Information Processing Systems 2015.
On a Joint Anomaly-Match Bayesian Aggregation Strategy (AM-BA) for Searching the Space of Source Hypotheses:
On Using Bayesian Sensor Reliability Models in Detecting Possible Spectrometer Faults and Failures:
On an Augmented PCA Approach for Incorporating new Background and Nuisance Source Fluctuations into Anomaly Model:
On Poisson Match Filtering for Boosting Detection Power of Low Count Photon Data from Inexpensive Wearable Spectrometers:
On Bayesian Aggregation to Boost Detection of Sources and Infer Source Properties:
Tandon, Prateek, Peter Huggins, Artur Dubrawski, Simon Labov, and Karl Nelson. Source Location via Bayesian Aggregation of Evidence with Mobile Sensors. Symposium on Radiation Measurements and Application 2012.
Tandon, Prateek, Peter Huggins, Artur Dubrawski, Jeff Schneider, Simon Labov, and Karl Nelson. Bayesian Aggregation for Radiation Source Detection. Pittsburgh Chapter of the American Statistical Association. Spring Banquet 2012.
On Bayesian Aggregation for Detecting Sources of Varying Intensity and Type:
On Adaptive Grid Bayesian Aggregation for Efficient Multi-Resolution Source Search
On using Poisson Principal Component Analysis to Boost Detection of Low Count and Weak Sources:
On a (team-built!) prototype robotic radiation sensing hardware backpack system useful for spectrometry controllable with an Android phone app:
On Use of Active Learning to Plan Routes for a Single Agent Robotic Vehicle to Detect a Radioactive Source:
On Use of Active Learning to Plan Routes for a Multi-agent Robot Team of Autonomous Vehicles to Detect a Mobile Radioactive Source in Traffic:
Thanks to the sponsors behind the ARI and DTRA projects that have funded my PhD work.