AI Sweden’s Edge Learning Lab is changing the way companies collaborate and innovate, and HPE infrastructure plays a key role.
Launched in 2021, AI Sweden’s Edge Learning Lab in Gothenburg works at the frontiers of one of the most exciting areas in AI. Where traditional training methods of AI models bring data to a central datacenter and operate on it there, Edge Learning systems start processing the data where it’s captured, with the capture device itself taking on a key role. In effect, each edge device builds its own segment of an AI training model, with all segments “fused” by a central aggregator, or one of the participating devices, to create a “supermodel” ready for deployment.
The result: less data needs to be transferred from the edge to the datacenter, and fewer resources will be needed to store the data. Furthermore, this enables models to be trained using data from a range of different sources, without the source data itself being shared.
The Edge Learning Lab stems from the work of AI Sweden, founded at Gothenburg’s Lindholmens Science Park in February 2019, formed, in the words of its Head of Data Factory, Mats Nordlund, “as a catalyst, to help accelerate the development and application of AI”.
Working both with Swedish companies and global corporations, AI Sweden acts as a sandbox where researchers, engineers and students from different organizations can get together, experiment, and share experiences and ideas. Crucially, where other AI research centers often focus on the development of new algorithms, AI Sweden focuses on the interplay between AI systems, hardware and infrastructure.
– You can tweak the infrastructure and that will have an effect on the algorithm and vice versa, Nordlund explains. From the very beginning, we put this in place together with partners like HPE, that brings the infrastructure to the Edge Lab, so that the algorithms and infrastructure can be developed simultaneously.
The impetus for the Edge Learning Lab came from AI Sweden’s automotive partners, as researchers wrestled with the challenges of transferring and processing the volume of data created by in-car sources.
– We had started to realize that the current direction on the infrastructure side – to just go with bigger and bigger datacenters and collect more and more data – works now but isn’t sustainable, notes Nordlund.
– We also anticipated restrictions on data transfer between countries. For example, Chinese data can’t transfer out of China, while GDPR in Europe also makes for restrictions.
The key trick here is the ability for partners to get together with their expertise and their equipment and mix it all together to accelerate learning and development.Mats Nordlund, Head of Data Factory, AI Sweden.
When automotive partners talked through these issues with those from other industries, all realized that they shared similar problems. For instance, hospitals could accelerate work on AI-powered diagnostics or treatment, if they could only pool their data, but patient privacy concerns and regulation make this impossible.
Research into Edge Learning promised a way forward, so a core group of AI Sweden partners including HPE, Volvo Cars and autonomous driving pioneer, Zenseact – joined forces to develop the Edge Learning Lab. The lab has all the compute, storage and network infrastructure in place for partners to experiment, rapidly building work environments for cutting-edge research.
– The key trick here is the ability for partners to get together with their expertise and their equipment and mix it all together to accelerate learning and development, says Nordlund. They can use the lab to solve what would otherwise be paradoxes for traditional AI training.
Having the equipment in place and the agreement between partners reduces the time involved in setting up a new project.
HPE has been a key player in the Edge Learning Lab from the outset, bringing equipment and expertise in edge computing, network infrastructure and decentralized swarm learning to help AI Sweden build the lab in just three months.
– HPE was one of the companies that saw the possibilities with this setup very early and became very strongly engaged, says Nordlund explaining that HPE provided “the know-how to set up an architecture for an Edge Learning system and the compute plan to power it”.
– We’re finding out how to build an edge learning system, how to make it converge quickly, how to get the required quality and how to update and maintain our models, he adds, because things change and you need to retrain your models.
Crucially, the lab’s infrastructure is based on the same general-purpose platforms and industry standard equipment that partners will be working with in the field, including HPE Edgeline Converged Edge Systems and edge-to-cloud architecture. This ensures that the results of experiments within the labs can be shared and replicated across industries.
AI Sweden has invented a new way to innovate with the help of strong partners in collaboration.
What’s more, by working with devices based on the latest AMD processor technology, the lab can push sustainability and energy efficiency without compromising on security or compute performance.
– Energy efficiency is always important, explains Nordlund, because the more energy efficient you are, the more application areas you can explore. It opens up more possibilities for edge learning.
Already, the lab is developing technology that can enable the Swedish forestry industry to collaborate on AI development to locate and monitor Spruce bark beetle infestations, without sharing sensitive business data. It’s also enabling satellite-based AI systems to train models in the satellites without occupying the limited bandwidth available for data transfer.
Perhaps most importantly, the Edge Learning lab is pioneering new ways in which companies from different industries, academia, and public sector bodies – even in different countries – can work together, where compliance or regulations would otherwise get in the way. The lab is attracting international collaboration to Sweden and transforming the way that collaborate.
– Both local and global organizations can learn really quickly and roll what they learn out in their innovation cycles, Nordlund explains.
He believes that AI Sweden has “invented a new way to innovate with the help of strong partners in collaboration”.