OctoAI: The next generation of high-performance edge AI for smart buildings


Current IoT (Internet of Things) solutions for buildings depend almost exclusively on cloud infrastructure and cloud-based services. In the OctoAI project, we are developing the next generation of high-performance Edge AI (Artificial Intelligence) for smart buildings. In OctoAI, we combine the concept of edge AI with user-centric energy services and test two edge-ready applications.

Starting point / motivation
Currently, the building stock in the EU remains energy-intensive and predominantly inefficient, accounting for 40% of final energy consumption and 36% of CO2 emissions. In order to increase the share of renewable energy and reduce energy consumption, future systems must have a high degree of flexibility and efficiency. Innovative Energy Services are built on bidirectional real-time interaction with buildings.

Innovative solutions are needed to generate, provide and analyze these large amounts of data. Internet of Things (IoT) technologies are the backbone and an enabler of these smart systems. Current IoT implementation depends almost exclusively on cloud infrastructure and cloud-based services. However, cloud-based services also have serious drawbacks: Reliability, trustworthiness, or security and privacy.

Contents and goals
Edge computing is an alternative IoT implementation and refers to computation taking place at the edge of networks; the "edge" is where end devices access the rest of the network. To realize the full potential of edge computing for smart buildings, "AI must be brought to the edge" of networks. The goal of OctoAI is to develop edge-ready AI models for smart buildings.

First, we define a set of qualitative goals by examining the tradeoff between a model's complexity, its performance, and its non-functional properties (e.g., latency and memory consumption). Centrally, we can measure our progress by defining a set of performance scenarios. In each scenario, we define a performance target for a target application on a target device.

Methodologically, OctoAI focuses on:

  • Scalable AI on the Edge. Resource-constrained edge devices do not provide the resources required to deploy state-of-the-art Deep Neural Networks (DNNs). Consequently, DNNs with a specific topology adapted to edge devices are required.
  • Transfer Learning on the Edge. Transfer learning refers to the reuse of a knowledge base captured in the form of an AI model for applications beyond the original use case.

Expected results

Outcomes of the OctoAI project are:

open algorithms and tools for edge-ready AI models for smart buildings and
a roadmap for edge computing in smart buildings.

Project duration

06/2022 - 05/2024

Project management

TU Graz - Institut for Software Technology

Project or cooperation partners


Contact Address

TU Graz
Institut for Software Technology
Gerald Schweiger
Inffeldgasse 16b
A-8010 Graz
E-mail: gerald.schweigernoSpam@tugraz.at
Web: www.tugraz.at/institute/ist/institute/