This talk discusses decentralised, context-aware Edge AI systems designed for bandwidth-limited, energy-constrained, and intermittently connected environments. It focuses on collaborative intelligence and resource sharing across multiple heterogeneous edge nodes.

Collaborative learning is enabled through peer-to-peer federated learning, inter-agent knowledge distillation, and distributed continual learning. Nodes incrementally update models using local data while exchanging knowledge to support specialisation, load balancing, and shared task execution. Continual and incremental learning are core to this process, allowing each node to adapt over time without central coordination.

This architecture supports efficient use of compute, memory, and sensing resources across devices such as UAVs, robots, embedded sensors, and wearables. Nodes can offload computation, share context, and coordinate responses based on availability, energy levels, or task priority.

The talk compares decentralised collaborative learning with traditional distributed frameworks, examining trade-offs in communication cost, scalability, model convergence, and autonomy. Real-world examples show how such systems operate as resilient, adaptive learning collectives in constrained environments.