Optimization for Neuromorphic Imaging and Digital Twins
Abstract
The project will procure hardware associated with Neuromorphic Imaging and Digital Twins at George Mason University (GMU). The hardware will help design new optimization algorithms. The project will prove to be a testbed for these new algorithms and subsequent deployment of this hardware in the field. Due to the large body of institutions in the Washington DC area, there is a tremendous need and opportunity to create such resources. The proposal identifies a large class of research problems, relevant to DoD, which cannot be handled using the existing or available hardware resources at GMU and the partner institutions. Two of these involve optimization problems arising in neuromorphic imaging and digital twins. In the first case, aircraft- and spacecraft-based cameras are subject to motion blur due to the high rate of vehicle speed. This blurring can make analysis difficult. Current space-based cameras also require a great deal of power and memory storage to operate in environments where power consumption and memory are extremely limited. Furthermore, traditional cameras cannot effectively operate in dark and high contrast environments, and similar challenges occur when tracking satellites or astronomical objects. Neuromorphic cameras emulate the human eye and can overcome all these challenges, which is extremely promising to AFOSR. To establish this, neuromorphic cameras will be acquired. New optimization algorithms will be designed to solve the underlying problems. The algorithms will further lead to improvement and deployment of neuromorphic technology. In the second case, one of the most pressing questions in science and engineering is how to combine physics-based models with data-driven models. This approach, when carried out for the entire complex system can be termed as a digital twin . These twins have shown promise to provide results in case the entire models are not available. An example is to develop adjoint based optimization strategies in determining weaknesses in structures, such as bridges, from sensor measurements. However, many challenges persist. The physical model could be very large, i.e. difficult to set up and to solve, and data may be of very high dimensionality. Significant uncertainty is expected at the interface of these two approaches. The new hardware targeted towards digital twins will help combine the physical constraints and data-based approaches such as machine learning. Note that either DoD does not currently have such hardware and algorithmic capabilities, or because of security issues it is almost impossible for students to access them. With the acquired hardware, the PIs will create an open environment which will be accessible to several neighboring institutions, including HBCUs. The PIs have identified several existing workshops and summer programs hosted by GMU to train researchers on this hardware. These events provide excellent opportunities for DoD to recruit the next generation of scientists. The hardware will also have direct impact on the currently funded research of the PIs. The resources are seen as a preparatory, educational, and developmental tool to make sure the technology related to neuromorphic chips and digital twins will be deployable in the field.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA95502410022
Entities
People
- Harbir Antil
Organizations
- Air Force Office of Scientific Research
- George Mason University
- United States Air Force