Machine Learning and Industrial IOT
by Chris Sachs, on Jun 7, 2017 2:37:15 PM
Announcing the Swim ESP™: Machine Learning at the IOT Edge
Enterprises and public sector organizations are drowning in a flood of sensor data from equipment, assets, suppliers, consumers, city data sources, and energy providers that contain hidden insights that have the potential to optimize production, transform efficiency, streamline flows of goods, vehicles and services, protect the environment and ensure safety. But finding them remains a huge challenge.
The tech community has responded in a predictable way – with more tech: Complex, big-data focused, cloud-hosted IOT solutions and new application development platforms. But these fail to recognize that organizations at the IOT edge – cities, utilities, manufacturers and suppliers - do not have the skills to develop new apps or implement and manage cloud hosted solutions. At its heart, this is not a technology problem – it’s a human problem.
Swim ESP™ addresses both the human problem and the tech challenge in one: Instead of new apps, complex infrastructure stacks and big-data, Swim uses powerful, efficient, real-time, distributed machine learning technology to discover insights automatically - at the IOT edge where streams of data arise. Swim finds hidden correlations and patterns without new application logic, using self-training edge learning algorithms on sensor data, and delivering a low-rate stream of insights in real-time that can be used for local decision making, offered to 3rd parties via APIs, or passed to planning applications for reporting.
Swim edge learning takes advantage of the price/performance benefits of Moore’s Law to deliver an affordable, easy to use, secure, and fault tolerant edge fabric that quickly delivers value in both brown- and green-field settings, on commodity edge devices, helping organizations quickly find hidden gems in their data:
- Swim implements distributed local learning: A digital twin of each real-world object learns from contextually relevant data streams, simplifying the learning problem.
- Swim edge learning is self-training. Its algorithms continually check their hypotheses against real world data, training and adjusting for over fitting as needed. This avoids the need to label training data and avoids any need for machine learning expertise in the OT environment.
- Swim runs on commodity edge hardware whose price/performance curve is relentlessly improved by Moore’s Law.
- Swim can learn as much on a device that costs a few hundred dollars at the edge, as a solution costing thousands of dollars per month, in the cloud.
- Swim uses learning to self-configure and manage, reducing IT/OT cost, training and complexity.
- Learned insights are available in real-time and in context, helping users or apps to quickly make smart, local decisions.
- By learning at the edge of the cloud, close to the systems that generate data, Swim delivers insights at a fraction of the cost.