VEDLIoT – Very Efficient Deep Learning in IoT

University scientists develop new hardware in EU-funded project to learn from experience

Autonomous vehicles or devices for asmart home are becoming increasingly complex in their requirements and processes. In the future, a new machine learning system should help to make the software and hardware of artificial intelligence applications more robust, more powerful and more energy-efficient. Scientists from the Computer Engineering group at the University of Osnabrück are working on such a solution together with eleven other partners. They are funded by the European Commission for three years in the project VEDLIoT (“Very Efficient Deep Learning in IoT”). It is being financed via the information and communication technologies funding line in the European Union’s Horizon 2020 program.

For the thematic background: In an intelligent home, a so-called “smart home”, residents will find devices designed to make their lives easier – for example, a refrigerator that can reorder groceries and communicate with the ovenat the same time. The devices and components are part of the Internet of Things (IoT). They are connected to a network and record, store, process and transmit data. IoT devices are also used in self-driving cars or industrial robotics.

In the VEDLIoT project, scientists from Germany, Poland, Portugal, Sweden and Switzerland are working together. Instead of classical methods, for example from the field of statistics, the international research team uses machine learning methods, such as deep learning. Artificial neural networks are used for this purpose. With the self-learning platform VEDLIoT, IoT devices should become more powerful and consume less energy at the same time. To achieve this, the researchers are developing a modular hardware platform that combines compact computing modules ofdifferent performance classes.

“We want to further optimize resource efficiency for artificial intelligence applications, which means increasing performance while minimizing energy requirements,” says Prof. Dr.-Ing. Mario Porrmann, head of theComputer Engineering group at the University of Osnabrück. “In doing so, we are pursuing new approaches in which the computer architecture is not fixed. The hardware of our computers will be capable of learning and will independently adapt to changing requirements during operation. Together with the AI Campus of the University of Osnabrück, the new project forms an ideal basis for our research work.”

“We want to further optimize resource efficiency for artificial intelligence applications, which means increasing performance while minimizing energy requirements,” says Prof. Dr.-Ing. Mario Porrmann, head of theComputer Engineering group at the University of Osnabrück. “In doing so, we are pursuing new approaches in which the computer architecture is not fixed. The hardware of our computers will be capable of learning and will independently adapt to changing requirements during operation. Together with the AI Campus of the University of Osnabrück, the new project forms an ideal basis for our research work.”

“Computer and IoT systems are becoming increasingly powerful. We can solve more challenging problems and advance automation to improve our quality of life,” says Prof. Dr.-Ing. Ulrich Rückert. He is the coordinator of the new VEDLIoT project and head of the Cognitronics and Sensor Systems Group at Bielefeld University. “But the amount of data collected and processed is huge and the computing power required is very high. In addition, the algorithms are often too complex to calculate solutions within a short time.” This is precisely where the new systems that the project team wants to develop come in.

In addition to universities and research institutes conducting research on artificial intelligence and the Internet of Things, companies are also involved in the European project; from Swedish start-up EmbeDL to German conglomerate Siemens. Other companies may also participate in the project. In addition to the existing applications in the automotive, automation and smart home sectors, at least ten more application examples are to be funded.

Author:

Mario Porrmann, Utz Lederbogen

Date:
15.07.2022

BACK TO project overview