Node Learning WorkingGroup

Chair: Prof Eiman Kanjo, Nottingham Trent University/Imperial College London

Co-Chair Dr. Wael Guibene, SiliconLabs

1. Introduction

1.1 Work Group Mission Statement
Advance the field of Decentralised and Collaborative Networked EdgeAI by transforming small, low-cost, resource-limited edgeAI devices into interconnected intelligent systems capable of learning, adapting, and making joint decisions.
Enable both autonomous and human-centred coordination through distributed computation, collaborative learning, and adaptive communication.
Promote research and innovation that integrate embedded systems, wireless networking, artificial intelligence, and sensor fusion to build the foundations for scalable, sustainable, and resilient edge AI ecosystems.

2. Key Definitions

• Decentralised AI: A framework where intelligence, computation, and learning are distributed across multiple nodes without central orchestration.
• Collaborative Learning: Multiple devices share features, parameters, and outputs to form a shared understanding and make joint decisions.
• Distributed Learning: Computation and learning spread across connected nodes that process local data and coordinate for global performance.
• Homogeneous Systems: Networks composed of identical nodes (e.g., identical microcontrollers or sensors) that distribute processing loads evenly.
• Heterogeneous Systems: Networks combining diverse devices with different sensors, processors, or memory capacities, contributing complementary capabilities.
• Resource Sharing: Allocation of processing power, memory, and sensing tasks across devices to overcome individual hardware constraints.
• Adaptive Infrastructure: A dynamic hardware–software environment that adjusts to changing datasets, communication conditions, and environmental variables.
• Federated and Decentralised Learning: Training and updating AI models collaboratively without transferring raw data between devices.
• Collective Decision-Making: Cooperative policy formation or prediction among distributed agents or devices through shared embeddings or decentralised reinforcement learning.
• Network size: Shared Intelligence across networks ranging from peer-to-peer setups of three nodes to thousands, using clustered or hierarchical coordination for sensing, learning, and decision-making.
• Continual Learning: Learning that allows a model to acquire new knowledge over time while retaining previously learned information. It adapts continuously to evolving data and tasks without full retraining.
• Incremental Learning: A stepwise learning process where new data or tasks are added gradually, updating the model without discarding existing knowledge. It focuses on efficiency and memory preservation.
• Continuum Learning: An extended form of continual learning where adaptation occurs seamlessly across a continuous stream of changing conditions, enabling stable, lifelong learning in dynamic environments.

1.3 Purpose
This work group aims to explore, develop, and validate decentralised AI frameworks that allow distributed nodes—whether sensors, drones, wearables, or embedded processors—to operate collectively with intelligence that emerges through collaboration.
It seeks to overcome the constraints of single-device AI by enabling computation, learning, and reasoning across a network of interconnected units. 

The group will focus on:

• Defining the concepts, principles, and terminology that shape decentralised and collaborative EdgeAI.
• Educating the community through reports, tutorials, and knowledge-sharing sessions.
• Consolidating a shared vision for decentralised intelligence across heterogeneous edge systems.
• Facilitating networking and discussion between academia, industry, and policy makers.
• Encouraging multidisciplinary collaboration between AI, embedded systems, wireless communication, and sensing experts.
• Promoting the development of benchmarking practices, shared infrastructure, and open testbeds by the wider research community.
• Supporting the creation of new standards and frameworks that guide reliable, scalable, and transparent EdgeAI research and deployment.

1.4 Deliverables

Educational Content

• Comprehensive introductory reports explaining decentralised AI and collaborative EdgeAI concepts.
• Tutorials on distributed learning, federated parameter exchange, and communication-efficient model synchronisation.
• Workshops demonstrating real-world implementations of decentralised learning on low-power microcontrollers and edge TPUs.
• Webinars introducing cross-domain methods connecting wireless protocols, AI model partitioning, and collaborative inference.

Community Engagement Activities

• Expert-led roundtables integrating academia, industry, and public-sector research labs.
• Panel discussions addressing open challenges in decentralised EdgeAI deployment, including data governance, interoperability, and model reliability.
• Collaborative hackathons to develop proof-of-concept systems combining heterogeneous devices such as UAVs, wearables, and ground robots.
• Networking events to foster long-term cooperation and exchange between researchers, developers, and hardware manufacturers.

Knowledge Dissemination

• Whitepapers detailing architectures for resource sharing, decentralised reinforcement learning, and collaborative perception.
• Technical reports evaluating federated convergence rates and communication bottlenecks in distributed environments.
• Regular updates and newsletters summarising the latest advances in decentralised AI, TinyML, and Edge computing.
• Documentation repositories containing open datasets, test configurations, and reproducible experiments.

Recognition and Amplification

• Case studies on collaborative decision-making between multi-modal EdgeAI nodes (e.g., combining audio, vision, and motion sensing).
• Success stories on distributed federated learning in agricultural, safety, disaster management, industrial, and environmental applications.
• Highlighting low-cost, low-energy prototypes demonstrating real-time decentralised inference and local adaptation.
• Member recognition through co-authored publications, demonstrations, and invited talks at academic and industrial conferences.

Practical Resources

• Frameworks for building distributed EdgeAI networks combining homogeneous and heterogeneous devices.
• Open-access simulation tools for evaluating communication reliability, model convergence, and latency under varying network conditions.
• Hardware assessment reports covering ARM Cortex-M7/M4, RISC-V, and AI accelerators such TPUs and NPU-enabled boards.
• Device assessment reports covering integrations calving, wearables, UAVs, ground robots.
• Practical guides on developing adaptive datasets that evolve to reflect new environmental or contextual inputs.
• Benchmarking tools to measure power efficiency, inference speed, and collaborative accuracy.

Future-Oriented Initiatives

• Construction of shared experimental testbeds connecting embedded AI, wireless mesh networks, and distributed sensors.
• Development of adaptive datasets capable of incremental updates to support continual learning across devices.
• Investigation into communication efficiency and routing reliability under non-ideal network conditions.
• Collaboration with policymakers and regulatory bodies to define standards for safe and ethical decentralised AI operation.
• Creation of low-energy, multi-purpose embedded systems that can be repurposed across domains such as healthcare, smart cities, and environmental monitoring.
• Development of educational and technical infrastructure that accelerates research and innovation across institutions.

Collaborative Publications

• Joint publications.
• Technical papers analysing experimental results from large-scale EdgeAI testbeds.
• Annual reviews summarising collective progress, challenges, and future directions for decentralised EdgeAI.
• Collaborative open-access datasets and reproducibility reports shared across the global research community.

---

2. Foundational Principles

2.1 Shared Vision & Goals
The work group shares a vision of decentralised intelligence that operates collaboratively, reliably, and sustainably.
The goal is to design systems where intelligence is distributed, computation is shared, and decisions emerge collectively rather than from a central authority.
This vision depends on cooperation between experts in embedded systems, wireless communications, AI model design, and sensing technologies.
Each discipline contributes to a common goal: scalable decentralised intelligence that adapts to real-world constraints.

2.2 Open Communication & Transparency
Promote transparency through continuous communication between members.
Use shared repositories for experiment data, source code, and benchmarking results.
Hold periodic open meetings and workshops where findings, challenges, and failures are documented and discussed.
Encourage reproducibility and community review for all major experiments and datasets.

---

3. Evaluation & Assessment

3.1 Metrics for Success

• Number of collaborative projects producing functional decentralised EdgeAI prototypes.
• Quantitative improvement in latency, bandwidth utilisation, and energy consumption compared with baseline centralised systems.
• Demonstrated scalability of decentralised learning across heterogeneous devices.
• Availability and adoption of public datasets, testbeds, and documentation produced by the group.
• Level of cross-disciplinary participation and long-term institutional collaboration.
• Academic and industrial citations of the group’s work in research papers, reports, and deployments.
• Integration of proposed frameworks into pilot projects or commercial systems demonstrating real-world value.
• Contribution to the development of standards or guidelines for decentralised and collaborative AI.