Specifically tailored for Critical Infrastructure Original Equipment Manufacturers (OEMs) and Device Manufacturers, the OctaiPipe platform serves as a catalyst for Data Scientists; offering unparalleled capabilities in building and orchestrating networks of intelligent Edge devices with Federated Learning (FL).
Unlocking the potential of connected IoT devices, OctaiPipe v2.1 streamlines the entire lifecycle for data scientists – from the initial stages of network setup and training to deployment, AI model fine-tuning, and continuous learning. Notably, this version introduces an exciting feature by adding support for edge gateway devices. This not only facilitates a reduction in hardware costs but also addresses the pressing concern of increased data privacy. Remarkably, all this is achieved without relying on high-performance embedded PCs.
The critical importance of OctaiPipe’s advancements becomes particularly evident when considering the diverse landscape of Critical Infrastructure companies, spanning energy, telecoms, civil engineering, and security. These sectors are becoming increasingly aware of the transformative power that AI and connected IoT devices can bring – from enhancing operational efficiency and sustainability to monitoring asset performance and ensuring network resilience.
Traditionally, the adoption of Cloud AI and connected devices in Critical Infrastructure environments has been hindered by the sector’s inherent characteristics. These environments are not only data-rich but also demand a high level of security. Intensive cloud resources requirements, security concerns, and network dependencies have posed significant challenges, limiting the sector’s utilisation of Cloud AI and connected devices.
Enter Federated Learning Operations (FL-Ops), a groundbreaking approach that OctaiPipe seamlessly integrates into its platform. FL-Ops empowers the deployment of AI at the edge and facilitates the management of distributed learning across a network of intelligent devices. Unlike conventional methods that involve moving data from Edge devices to the cloud for AI algorithm training – with all the associated data storage costs and security apprehensions – Federated Learning (FL) takes a decentralized route.
This not only revolutionizes the way AI is deployed in Critical Infrastructure but also alleviates the challenges posed by data-intensive operations, security requirements, and network dependencies. And in our latest update, OctaiPipe is harnessing its Federated Learning power on even smaller devices, further reducing compute and power requirements.
For an explanation on Federated Learning for Edge AI for IoT, watch OctaiPipe’s video below.
OctaiPipe v2.1 updates
Version 2.1 places a primary emphasis on edge gateway devices with a lower specification than traditional embedded PCs. This allows the benefits of federated learning to be delivered closer to the edge using hardware that is already part of the data collection infrastructure.
What else in new in OctaiPipe version 2.1?
Federated Scikit Learn models on 32bit ARM devices
Scikit learn models are often less resource intensive than deep learning approaches, so can be beneficial for solving problems like motor anomaly detection using low-powered 32bit ARM devices, often found in edge router devices. Version 2.1 includes support for federated learning with Scikit learn models so that when needed, anomalies can be detected by combining the training results from multiple locations, without compromising the privacy of data at each location.
Reduced footprint OctaiPipe learning and inference components
In order to support both training and prediction of ML models on edge devices with a lower specification than typical industrial PCs, we have produced a resource-optimised version of OctaiPipe targeted at these devices. This has allowed us to reduce the RAM and storage requirements on the edge device by up to 500MB.
Robust and secure OctaiPipe edge device client
Version 2.1 introduces a new process for securely installing the OctaiPipe client on each edge device with a unique on-time passkey for that device. This ensures that only approved devices can interact with the OctaiPipe portal to join federated learning experiments and receive the resulting trained models.
Support for reading and writing data from MQTT message brokers
Many edge devices receive or transmit data via MQTT buses such as Eclipse Mosquitto™. Users can now create OctaiPipe custom steps to read and write from MQTT message brokers to extract aggregate and forward data as or train models at the edge directly from incoming data streams.
Continued improvements to OctaiPipe automated platform security tests
OctaiPipe development processes includes automatic security and penetration testing with updates during each release to ensure intrusion resistance is maintained.
Provide more information about FL experiments and deployments in developer and management UI
More complete and detailed information is now provided in both the management and development UI for experiments and deployments so users can easily track progress of federated learning and delivery of the resulting models.