EdgeMind: A Self-Evolving AI Framework for Distributed Intelligence in IoT Ecosystems

Intelligence in Internet of Things (IoT) ecosystems is largely centralized and limited in its capacity to adapt, learn new tasks quickly, and optimize its operation. EdgeMind offers a framework for self-evolving distributed intelligence, enabling a high level of autonomy for intelligent applications. Each application can be decomposed into dedicated components along the edges of the cloud-edge continuum, acting on local data and collaborating with others for coordination and knowledge sharing. A diverse range of learning and adaptation mechanisms is supported to address these tasks and drive continual improvement of intelligent components. More broadly, EdgeMind provides an operational architecture to support intelligence at multiple levels of autonomy; the focus here is on applications that require a significant level of independence.

A meta-architectural approach is adopted, allowing independent design and development of components. Novelty is embedded within the module-specific learning loops that enable EdgeMind to meet the intelligence-oriented research directions set forth in recent reports by the UK Government Office for Science, AI Council and Defence and Security Accelerator. The design welcomes adaptation and optimization of operation (including the profile of computation and communication resources) through self-optimization, continual learning and meta-learning. EdgeMind is therefore positioned to go beyond established paradigms such as federated learning by addressing the learning and adaptation needs of applications across the full diversity of the AI/ML lifecycle and operational continuum.