Mats points out that data is often born on the edge, for example in devices that may be in your pocket, or equipment in a manufacturing system, or sensors in your car. Using traditional methods, the data is transferred to a central location in order to train the algorithms. Transferring a lot of data to a central location has a lot of challenges:
“Consider the data generated at a hospital. This is very valuable, but in many cases it cannot be shared with other hospitals for privacy reasons. In other applications, we collect huge amounts of data, but the transfer and storage costs make it challenging to move all this data to a cloud or a central storage and compute location. Both these scenarios beg the simple question - can we have the cake and eat it too? Decentralized AI may be the answer. How do we share knowledge and insights without transferring the data.”
Ather, who is taking a leading research role in AI Sweden’s newly formed strategic program in Decentralized AI, says that this program marks an important step to address these challenges.
“In order to fully accelerate the use of AI we need to operationalize the outcomes and learnings from individual, collaborative projects so that they can be used at a larger scale. AI Sweden is therefore moving from managing a portfolio of individual projects to developing a number of dedicated strategic programs, one of these is Decentralized AI. These programs will leverage each other and drive the use of results, learnings, data, methods, and best practices to many partners. Decentralized AI is in its infancy and at AI Sweden we want to become world leaders in developing this technology based on applied research conducted in partnership with our partners who will benefit from the technology.”
Mats explains that one of the spearhead initiatives within the program is the Edge Lab that was opened earlier this year. The Edge Lab is part of the Data Factory testbed facilities and lets industry partners come together to innovate around edge learning:
“With partners such as Volvo Cars, Ericsson, HPE, Zenseact and CGit providing expertise and infrastructure, the Edge Lab is at the absolute forefront of edge learning. In addition to supporting organizations in Sweden to benefit from this new technology, we also invite leading international industry and academic partners to this lab.”
Concluding, Ather underlines that the Decentralized AI program addresses real-world challenges. “It has a strong commitment from partners in both the public and private sector. The unique setup and broad partnership enable us to contribute to the field at a global level. From an applied research perspective, we will focus on identifying solutions to common challenges, that have the potential to accelerate the use of decentralized machine learning for increased competitiveness in many different organizations and sectors.”