Federated Auto-Meta-Ensemble Learning Framework for AI-Enabled Military Operations

Abstract

One of the promises of AI in the military domain that seems to guarantee its adoption is its broad applicability. In a military context, the potential for AI is present in all operational domains (i.e., land, sea, air, space, and cyber-space) and all levels of warfare (i.e., political, strategic, operational, and tactical). However, despite the potential, the convergence between needs and AI technological advances is still not optimal, especially in supervised machine learning for military applications. Training supervised machine learning models requires a large amount of up-to-date data, often unavailable or difficult to produce by one organization. An excellent way to tackle this challenge is federated learning by designing a data pipeline collaboratively. This mechanism is based on implementing a single universal model for all users, trained using decentralized data. Furthermore, this federated model ensures the privacy and protection of sensitive information managed by each entity. However, this process raises severe objections to the effectiveness and generalizability of the universal federated model. Usually, each machine learning algorithm shows sensitivity in managing the available data and revealing the complex relationships that characterize them, so the forecast has some severe biases. This paper proposes a holistic federated learning approach to address the above problem. It is a Federated Auto-Meta-Ensemble Learning (FAMEL) framework. FAMEL, for each user of the federation, automatically creates the most appropriate algorithm with the optimal hyperparameters that apply to the available data in its possession. The optimal model of each federal user is used to create an ensemble learning model. Hence, each user has an up-to-date, highly accurate model without exposing personal data in the federation. As it turns out experimentally, this ensemble model offers better predictability and stability. Its overall behavior smoothens noise while reducing the risk of a wrong choice resulting from under-sampling.