Presented By O’Reilly and Intel AI
Put AI to Work
April 29-30, 2018: Training
April 30-May 2, 2018: Tutorials & Conference
New York, NY

Using machine learning to enhance activity-based intelligence

Jamie Irza (Raytheon)
1:45pm–2:25pm Tuesday, May 1, 2018
Implementing AI
Location: Concourse A
Average rating: ***..
(3.00, 1 rating)

Who is this presentation for?

  • Sensor system engineers and AI/ML practioners

Prerequisite knowledge

  • A basic understanding of machine learning

What you'll learn

  • Learn how machine learning can be used to implement the methodologies of activity-based intelligence (ABI)
  • Discover how ABI can be used to detect anomalous flight patterns and how ABI, in conjunction with machine learning, can enhance the performance of traditional airspace sensors such as radars

Description

Knowing the difference between “normal” and what constitutes an anomaly is crucial to the successful operation of many types of systems. Perhaps the most widely known example of anomaly detection is the capability to detect credit card fraud. The key to creating an anomaly detection system that has a low false-alarm rate (false positives) and a high probability of detection is the ability to utilize a model that accurately portrays “normal behavior.” For systems involving humans, capturing normal behavior can be thought of as understanding normal patterns of life.

Activity-based intelligence (ABI) is the art and science of understanding normal patterns of life to enhance the ability of a system to detect anomalous behavior. ABI has traditionally been a manual, time-consuming effort requiring a human analyst to assimilate information from a variety of sources. This has resulted in analyst overload and untimely delivery of anomaly detections.

Many aspects of ABI and anomaly detection can benefit from the application of AI and machine learning techniques. These techniques can provide as-good-as or, in some cases, better performance than a human analyst or operator. Jamie Irza demonstrates how machine learning can be used to implement ABI for detecting threatening behavior from unmanned aerial systems, commonly known as drones. Of particular importance is the ability to use historical air traffic patterns (patterns of life) to alert an air traffic operator to the potential presence of drones operating in manned airspace or near sensitive areas. Using a variety of machine learning algorithms to implement ABI, a radar system can exhibit enhanced detection of drones flying “low and slow” in clutter-filled regions. Jaime offers comparison of several popular ML algorithms that address this scenario and the requirements for training datasets.

Photo of Jamie Irza

Jamie Irza

Raytheon

Jamie Irza is a senior principal systems engineer at Raytheon Integrated Defense Systems, where her recent work has been focused on the application of machine learning techniques to implement activity-based intelligence (ABI) algorithms to enhance the performance of sensors such as radars and imaging systems. Jaime’s technical specialization includes systems engineering, signal processing, and machine learning. Jamie holds a BSEET with a minor in mathematics from Roger Williams University and an MSEE from the University of Rhode Island.