Fast and Reliable Object Identification via Guaranteed Learning on Neural Networks
Abstract
Army networks and systems are generating and collecting vast amounts of data with different modalities, sampling rates, and accuracies continually. The obtained data often suffer from quality issues due to unreliable or defective sensors, poor atmospheric and lighting conditions, and cyber data attacks from malicious intruders. The processing and information extraction from these massive amounts of data for real-time decision making becomes exceptionally challenging. Anomaly detection and object identification are two tasks at the heart of situational awareness for Army applications. Anomaly detection sends alarms for abnormal events, and object identification locates human objects and predicts their movements. Compared with the classification and identification methods in civilian applications, the methods in a military application must have much higher reliability, as a minor error can potentially lead to the loss of lives of military personnel. Therefore, the deployed identification methods need to have analytical performance guarantees rather than numerical success only. This proposal will develop computationally efficient anomaly detection and object identification methods from large amounts of data. One distinctive feature of developed methods in this proposal is that they are accompanied by analytical performance guarantees which enable reliable implementation in military operations.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 04, 2023
- Accession Number
- AD1212674
Entities
People
- Meng Wang
Organizations
- Rensselaer Polytechnic Institute