Application of riblets to control adverse pressure gradient turbulent boundary layer
Abstract
Goal- The primary objective of this proposal is to develop a scalable multi-agent AI system that enhances its performance through experience, demonstrates the ability to solve previously un-encountered situations, exhibits robustness in the face of agents failures, and minimizes learning time, communication, and energy requirements. Motivation- While the current deep learning methodology is immensely successful, it faces significant scalability challenges. This is primarily due to its need for extensive data and substantial energy consumption. The large data collection may lead to ethical concerns as well. But worse, after this painstaking effort, existing AI technologies remain static and struggle to adapt to dynamic environments or handle less predictable conditions. We firmly believe that it s time for a transformative shift, especially considering the recent advancements in the state of the art and our own achievements in continual learning and curriculum learning. Novelty- The outcomes of our research are poised to revolutionize the life cycle of machine learning systems. By reducing dependence on vast datasets and lowering energy consumption, we aim to enhance the adaptability and longevity of AI technology. Our innovations encompass several key aspects- (1) Adaptivity in Training- We will develop an adaptive learning approach that can seamlessly switch between different training programs, drawing inspiration from the human ability to employ diverse learning methods. This approach is anticipated to conserve resources and bolster resilience. (2) Multi-Agent Lifelong Learning- We will design a collaborative community of lifelong learners. These agents will engage in communication to both enhance immediate decision-making and facilitate continuous learning, and as the agents improve, they require less and less communication. (3) Optimized Cooperative Lifelong Multi-Agent System- Our system will operate on two levels of continual learning. The low level will rely on environmental cues and guidance from a teacher, while the high level will incorporate knowledge obtained from peer agents. These interactions with both teachers and peers will enable the system to be updatable, while also expediting the adaptation process, reducing energy and data requirements, and enhancing overall system reliability. Significance- Our study holds profound implications. It will establish the mathematical and algorithmic foundations for airborne heterogeneous autonomous systems capable of efficient communication to support each other under ongoing demands and communication constraints. These foundations will also shed light on rapidly deploying upgraded vehicles and systems that lost certain functionalities (e.g., operation until reinforcements arrive).
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA86552417008
Entities
People
- Amirreza Rouhi
Organizations
- Air Force Office of Scientific Research
- Nottingham Trent University
- United States Air Force