IoBT CRA
IoBT Innovations in Edge Efficiency
IoBT Improving Edge Efficiency:
Efficiency: This video demonstrates how multiple IoBT innovations jointly contribute significant reductions in sensor-to-decision latency by pushing AI-based analytics (needed for fine-grained target classification) to the point of observation.
IoBT Rapid Network Synthesis:
Efficiency: This video demonstrates IoBT rapid network synthesis algorithms that significantly improve the scalability of network synthesis by finding linear/polynomial approximations to the original intractable (i.e., NP-Hard) problem while guaranteeing near-optimality.
IoBT Fast Anomaly Detection:
Efficiency: The video demonstrates a fast anomaly detection system that exploits the foundations of quickest change detection, developed in the IoBT project, to significantly reduce detection latency.
IoBT Compressive Offloading:
Efficiency: The video demonstrates improvements in compression of communication (between sensors and computing nodes that run downstream analytics) that exploit mission goals and the nature of analytics to significantly improve compression ratio.
IoBT Innovations in Resiliency
IoBT Trustworthy Machine Learning with Surprise Detection:
Resilience: This video demonstrates a capability for identifying when confidence in neural network outputs should be reduced because the input is somehow anomalous (e.g., comes from a different distribution compared to training data, includes a previously unseen class, or presents objects in a very unlikely context).
IoBT Risk Aware (Task) Placement:
Resilience: This video demonstrates an algorithm for risk-aware placement of computation at the distributed tactical edge. It replicates sensitive computation in a manner responsive to risk level, such that failures (of one or more nodes) do not disrupt critical computational and decision services.
IoBT Resilient Localization:
Resilience: This video demonstrates a resilient localization algorithm that maintains correct position estimation in the presence of compromised sensors that attempt to jeopardize localization accuracy.
IoBT Innovations in Intelligence at the Point of Need
- Example #1: Multimodal Sensing for Human Monitoring: Video
- Example #2: Vital Sign Assessment (for Survivor Triage) Using Multimodal Sensing: Video
- Example #3: Unconventional Sensing (Audio Sensing with Accelerometers): Video
- Example #4: Exposing Hidden Sensors: Video
- Example #5: Distributed Neural Network Partioning with CLIO: Video
- Example #6: URSA Bench, A Benchmarking System for Uncertainty, Robustness, Speed, and Accuracy: Video
IoBT Multimodal Sensing for Human Monitoring:
Multimodal Sensing: This video demonstrates a range of radio-frequency-based multimodal sensing capabilities that utilize commodity radios as sensing devices for a variety of human target detection and state estimation applications.
IoBT Vital Sign Assessment (for Survivor Triage) Using Multimodal Sensing:
Multimodal Sensing: The video demonstrates a capability for detecting survivors in busy environments remotely by detecting their vital signs using multimodal radio-frequency sensing.
IoBT Unconventional Sensing (Audio Sensing with Accelerometers):
Multimodal Sensing: This video demonstrates a novel neural network architecture with an “alias unfolding” module that leverages sparsity to reconstruct high-frequency signals from low-frequency sensor recordings (i.e., recordings sampled below the Nyquist frequency of the original signal). It is used to reconstruct recognizable sounds from their accelerometer signatures.
IoBT Exposing Hidden Sensors:
Multimodal Sensing: This video illustrates an IoBT capability for exposing hidden sensors by stimulating the (potentially present) sensor to induce a detectable electronic signature that exposes it.
IoBT Distributed Neural Network Partitioning with CLIO:
Distributed Edge Analytics: This video demonstrates distributed load sharing of edge analytics via neural network partitioning among heterogeneous nodes using the “Cloud-IoT” model-partitioning architecture, called CLIO. The architecture allows collaborative execution of (neural-network-based) tasks, where no single node is sufficiently well-equipped to carry out the task individually.
IoBT URSA Bench: A Benchmarking System for Uncertainty, Robustness, Speed, and Accuracy:
Uncertainty Analysis: The video demonstrates URSA Bench: A novel benchmarking system for neural network uncertainty, robustness, speed, and accuracy that incorporates advantages of Bayesian Learning.
IoBT Experimental Infrastrcuture
IoBT Resilient Publish-Subscribe Architecture:
Experimentation Infrastructure: This video describes a resilient publish-subscribe protocol implementation for the IoBT testbed.
IoBT Multimodal Data Collection Platform:
Experimentation Infrastructure: This video demonstrates a sensing platform for IoBT multimodal sensing experiments.
IoBT Multi-Sensor Array (MSA) Testbed:
Experimentation Infrastructure: This video describes the MSA experimental facility used by IoBT.