Machine Learning & Sensor Management for High Throughput Screening
This project is investigating the development of automated explosive detection and classification algorithms for high throughput screening. This is critical both in portal systems, where high throughput requires significant automated decision support, and in stand-off systems where the proliferation of multimodal data can overwhelm human interpretation. The project’s fundamental assumption is that it is too slow or costly to collect full sensor data on every object of interest, either for training, or during real-time operation. As a consequence, there are several important problems to address. In training, one needs to select which data will be used to train the decision algorithms in order to achieve robust performance. This is a problem known as active learning. In the real-time phase, one needs to use a hierarchy of sensing and classification strategies, based on relatively inexpensive early warning sensors, and adaptively select subsequent sensor measurements in order to arrive rapidly at an accurate classification decision. The long- range impact of this research will be the development of adaptive, high throughput screening algorithms for different combinations of sensing modalities that exhibit improved sensitivity/specificity.
The long-range impact of this research will be the development of adaptive, high throughput screening algorithms for different combinations of sensing modalities that exhibit improved sensitivity and specificity.- from F3-C Progress Report
Students Currently Involved in Project
- Kirill Trapeznikov
- D. Motamed Vaziri
- Joe Wang