SBIR/STTR Award attributes
This effort aims to develop interpretable and reliable machine learning methods that address the challenge of deriving explanations of autonomous decision-making behavior. In particular, the proposed effort focuses on the challenges inherent in interactions between humans and intelligent machines, where transparency and trust are essential to facilitate successful human-machine teaming. This effort, based on deep reinforcement learning, will address the fact that autonomous decision-making agents can affect future states based on their current actions, as well as the challenge of reasoning over long-term human-machine collaborative objectives of the underlying mission. The resulting explainable system enables better understanding of learning outcomes, and can also help develop more effective machine learning algorithms. The developed systems can be applied to military scenarios by enabling human-interpretable behavior explanations in human-in-the-loop decision processes. These systems may also be deployed in commercial applications such as autonomous driving or energy management systems, where high-stakes decisions require transparency and traceability. The proposed explainable machine learning techniques can also be implemented in heavily regulated areas, such as healthcare or financial systems, where stringent interpretability and accountability are required. The objective of this effort is to conduct a feasibility study and validate prototype concepts for future development and integration.
