Machine learning for AI is vital to automating, scaling, and maintaining the smart critical
infrastructures and systems on which society increasingly relies to keep the world safe, secure, and sustainable.
Machine learning requires enormous scale to meet the performance requirements of critical systems. This is acutely the case in distributed IoT/Edge systems where data is typically in very high total volume but the meaningful events, states, and behaviours to be learned are scarce. Paradoxically, the sheer scale and magnitude of distributed devices, data and processing which is required to achieve performance that can be trusted and relied upon, currently creates unacceptable risks and prohibitive costs. Both are vast and are measured in the tens of millions of pounds per use case.
Today, the cause of this is the outdated centralized machine learning paradigm, entirely reliant on complex, vulnerable, and costly centralized cloud and network solutions which are simply not designed or optimised for tasks using decentralized data and devices at real world scale.Locked and segregated by physical, legal, and security-based data silos, constrained by network, and almost totally dependent on cloud compute, the resulting unacceptable privacy and security risks of data access and sharing, economic unviability of data transfer and model training, and insufficient resiliency of operation leads to massive AI underperformance or prohibits it entirely.
The fact is that today, using current systems, AI for IoT simply cannot be sufficiently trusted or relied upon.
The impact is that many otherwise highly valuable use cases are unviable, and so are
either not scaled or left dormant and unsolved, while edge computing remains vastly
underutilized.
For distributed critical systems, a new, fit for purpose and optimised edge AI infrastructure is required, designed specifically to solve these problems which currently prevent the use of AI in IoT at scale. OctaiPipe solves this by delivering to market the first – and best in class – Federated Edge AI for IoT platform for smart critical infrastructure and systems. It provides a dedicated, fit for purpose alternative to the centralized cloud for AI engineers to build, deploy and manage continuously and collaboratively learning systems that can be trusted to intelligently automate IoT/edge powered solutions. Applications include machine learning for robotics/mechatronics/intelligent automation, predictive maintenance, process optimisation, digital twin, quality inspection, supply/demand forecasting, energy distribution modelling, health and safety monitoring, cyber security and more.
Core to the solution is our innovation in federated machine learning for edge applications, which rather than centralising data to the cloud for training, trains models on networks of collaborating devices that combine learning outcomes as a group to solve complex problems. This approach, and associated federated model ops infrastructure improves model performance 2x, guarantees data security and privacy, learns 90% faster, reduces cloud compute costs x100-x1000, delivers >50x network cost efficiency, and hardens system resiliency to cyber-attack against AI and natural degradation. The result is 5-10x ROI for machine learning problems compared traditional methods.
The highly privacy and security preserving nature of federated edge AI and its associated cost efficiencies, mean that it is applied in 3 core contexts. Single organisations with distributed structures and systems seeking to leverage cost efficiencies. Organisations looking to overcome data sharing barriers to improve models by training at scale across other entities e.g. an OEM improving products by securely training models across customers while maintaining privacy.
Organisations collaborating to solve common objectives without the need or risk of sharing data directly with each other e.g. FMCG blue chips collaborating to learn health and safety risk models across organisations and their sites, while maintaining data privacy and GDPR compliance.
To the AI engineer, OctaiPipe’s platform UI renders the complex, simple. Like never before, a single data scientist can now train, deploy and manage trusted AI to whole, vast IoT/Edge systems in critical systems to keep people and machines operating efficiently, safely and securely. OctiaPipe radically reduces the costs and risks of machine learning at scale. The impact is that with acceptably low risks and costs, AI for IoT can now be trusted and scaled to solve the previously unsolvable problems preventing us from keeping critical systems – and those that rely on them – safe, secure, and sustainable.