arXiv:2610.01647v1 Announce Type: new Abstract: In this article, we present our vision and recent advancements toward creating a decentralized system capable of learning from real-time data within buildings to support sustainable and privacy-preserving smart environments. Our approach promotes the concept of the building itself as the data center, aligning with the principles of edge computing to
arXiv:2610.01647v1, submitted on October 1, 2026, by Christophe Cérin, Mamadou Sow, and Frédéric Andrès, proposes a decentralized cloud-fog-edge architecture for smart buildings that treats the building itself as the data center. This approach prioritizes privacy, data sovereignty, and energy efficiency by enabling in-situ learning on low-power microcontrollers (e.g., Arduino/ESP32) rather than relying on external cloud infrastructure.
The system utilizes a lightweight, Kubernetes-like orchestration framework called KOptim, integrated with FIWARE for dynamic Service Level Agreement (SLA) management. Key technical implementations include: Online Machine Learning: Deployment of algorithms like K-means clustering and GHA dimension reduction directly on edge devices to handle real-time sensor data (CO2, temperature) and concept drift. Hybrid Architecture: Balancing workloads across cloud (centralized management), fog (building-level servers for localized AI and conflict negotiation), and edge (individual sensors and microcontrollers). * Humanitarian Focus: Designed for resource-constrained environments, the system supports federated learning and decentralized decision-making to enhance resilience and autonomy in underserved communities.
The paper presents a vision for a cloud–fog–edge architecture in which the smart building itself functions as a distributed data center. Rather than treating the building as a passive source of telemetry that must be shipped to a centralized cloud, the authors advocate processing and learning from real-time building data as close to the sensors, actuators, and subsystems as possible. The central idea is to use edge and fog layers for low-latency inference, local coordination, and data aggregation, while reserving cloud resources for heavier analytics, model management, or cross-building coordination where appropriate.
Its main contribution is a building-centric, decentralized framework for privacy-preserving and sustainability-oriented smart environments. The approach is aligned with edge-computing principles: keep sensitive or high-volume data local, reduce dependence on external networks, and enable adaptive control of building systems from real-time signals. This suggests a shift from cloud-only smart-building analytics toward an architecture that can support on-site learning, faster response to occupancy or energy-demand changes, and stronger data-minimization practices.
The work matters because it addresses several practical limits of conventional smart-building systems. Cloud-centric designs can introduce latency, bandwidth costs, privacy exposure, and operational fragility, especially when buildings generate continuous streams of environmental, occupancy, and energy data. By positioning the building as an active computational and data-processing entity, the paper offers a roadmap for smarter, more resilient, and more sustainable building operations—particularly in settings where privacy, energy efficiency, and real-time adaptability are central design constraints.