CCICADA Seminar Series in Homeland Security-2/16/17
CCICADA Seminar Series in Homeland Security. Every other Thursday from 12:15 pm to 1:00 pm
** ANNOUNCEMENT **
Command, Control and Interoperability Center for Advanced Data Analysis – CCICADA Research Group Lab Meeting – February 16, 2017
FEATURED SPEAKER:
Simon Streltsov, LongShortWay Inc., Boston MA
Machine Learning Technologies for Wide Area Search
Wide area sensors, such as Wide Area Motion Imagery (WAMI), Ground Moving Target Indicator (GMTI) radars, satellite imagery, provide information about movements in large areas. These sensors play important role in ground operations, border protection, and maritime surveillance. Due to relatively low spatial resolution of the data, detection and tracking in such data is challenging not only to computer algorithms but to human analysts as well [Ling 2011].
We will describe machine learning technologies for target detection and characterization, activity detection, and tracking in wide area sensor data. We address challenges in ground truth collection through collaboration with analysts and co-collects from multiple sensors. We will also describe development of two new learning technologies that address challenges of providing robust decision rules for such practical applications with limited amount of ground truth.
Feature Clustering (FC) groups features into separate clusters, based on the feature inter-correlations. Thus, FC enables partitioning of the classification problem into approximately independent sub-problems. The feature space partitioning enables solving each sub-problem independently and then fusing results into one final classifier [Kuncheva 2014].
In the Learning Under Privileged Information (LUPI) paradigm [Vapnik 2009, 2015; Streltsov 2014; Ilin 2016], a Teacher supplies students with additional information that is available during training only. LUPI algorithms allow better generalization from a small number of training samples.
This is joint work with Ilya Muchnik, Yuri Goncharov, Alan Gove, Steve Bento.
Simon Streltsov is the President of LongShortWay Inc. He had previously worked at Alphatech Inc. and Mercury Computers. Dr. Streltsov received PhD in Manufacturing Engineering from Boston University in 1996. LongShortWay Inc. specializes in developing machine learning applications for intelligence and surveillance applications for US Air Force, Air Force Research Laboratory (AFRL), Army, and DARPA.
REFERENCES
1. H. Ling, Y. Wu, E. Blasch, G. Chen, H. Lang and L. Bai, “Evaluation of visual tracking in extremely low frame rate wide area motion imagery,” in Proc. IEEE International Conference on Information Fusion, Chicago, 2011.
2. V.Vapnik and A.Vashist, “A new learning paradigm: learning using privileged information,” Neural Networks 22, pp.544-557, 2009.
3. Vladimir Vapnik, Rauf Izmailov “Learning Using Privileged Information: Similarity Control and Knowledge Transfer”, JMLR 16(Sep):2023−2049, 2015.
4. C. Curtis,J. Patrick “Medium Area Motion Imagery (MAMI) Truth Tracking Training,” 2014
5. Ludmila I. Kuncheva Combining Pattern Classifiers: Methods and Algorithms 2nd Ed, 2014.
6. Roman Ilin, Rauf Izmailov, Yuri Goncharov, Simon Streltsov Fusion of privileged features for efficient classifier training Information, 19th International Conference on Fusion, 2016.
7. Simon Streltsov, Alan Gove, Ilya Muchnik, Kirill Trapeznikov, Venikatesh Saligrama, David Castanon, Akshay Vashist New Learning Technologies for Exploitation of Layered Sensor Data, AFRL report, April 2014 (joint work with Boston University and Vencore Labs/ACS).
Date/Time:
Thursday, February 16, 2017/12:15pm to 1:oopm
Location:
Computing Research & Education Building (CoRE), Busch Campus, Rutgers, the State University of NJ
4th Floor Conference Room – 433
Leave a comment