Traditional AI is reaching its limits in the distributed world of Internet of Things (IoT). Data privacy and security is a key concern for IoT technology, particularly where that data is personal or security sensitive. The number of IoT devices and enabled machines is growing exponentially, estimated to reach 26Bn globally by 2030. Creating platforms optimsed to maximise private and secure machine learning that can be trusted, is an urgent priority.
In response, OctaiPipe has a built a distributed, secure, auditable, and privacy preserving machine learning platform, targeting deployments to IoT and edge devices. OctaiPipe combines federated learning, automated ML, and fault-tolerant Edge MLOps to automate the entire ML lifecycle on device or connected machine. This delivers a unique market offering connecting the customer to AIoT that is more private, secure, efficient, and autonomous.
This project seeks to undertake a feasibility study to develop a consortium of hardware and software partners, as well as sector specific technology consumers, to further the development and testing of trustworthy Edge AIoT. This development will follow four main axes;
1) Adversarial fortification: Identify and develop processes to protect FL against poisoning attacks
2) Privacy-Accuracy trade off: Identify and develop processes to automatically optimise model parameters to balance privacy and accuracy performance on edge IoT devices.
3) Explainable Edge MLOps: Develop algorithms that explain the output of quantized compressed ML models typically deployed to edge devices
4) Auditable Edge MLOps: Identify and establish transparent edge MLOps processes that are easily auditable by end users without sacrificing privacy and security.
This project will promote rapid end-to-end adoption of OctaiPipe at scale. Doing so will solve the technological challenges to trustworthy AI for IoT. This will reduce time to impact of AI software that is more adoptable by virtue of being private, secure, interpretable, and accountable, further boosting the UK as a leader in data and AI technology, leading to significant and rapid economic benefits. In summary, this project aims to address vulnerabilities in machine learning for IoT with specific focus on FL and addresses the fundamental challenges of socially responsible AI adoption into society.
