João Folhadela from the Institute for AI Safety and Security presents the Next Generation Reservoir Computing (NG-RC) approach, which was implemented specifically for FPGAs.
The DLR Institute for AI Safety and Security has developed an innovative Next Generation Reservoir Computing (NG-RC) approach for FPGAs
The new technology requires significantly fewer computing resources than conventional deep neural networks with comparable results
Application potential in robotics, IoT applications and resource-limited environments
Presentation at the renowned FPGA Developers' Forum at CERN
DLR presents resource-saving AI technology at international FPGA forum
Artificial intelligence is conquering industry, but the most popular AI tools such as deep neural networks are often resource-intensive. The DLR Institute for AI Safety and Security is therefore researching innovative alternatives to achieve comparable results with significantly less complexity. At the FPGA Developers' Forum at CERN, the researchers presented a Next Generation Reservoir Computing (NG-RC) approach implemented specifically for FPGAs.
Breakthrough in the efficiency of AI models
The method developed by DLR impressively demonstrates how efficient AI models can outperform conventional neural networks in predicting complex systems - while requiring only a fraction of the resources. Field-Programmable Gate Arrays (FPGAs) offer an ideal platform for this type of resource-efficient AI implementation. The Next Generation Reservoir Computing approach differs fundamentally from traditional neural networks. While conventional deep learning approaches have to train all network parameters, reservoir computing uses a hardwired, randomly generated network and only trains the output layer. This significantly reduces the training effort and makes the technology particularly attractive for resource-limited applications.
Promising fields of application
The research results open up exciting new prospects for use in robotics, Internet-of-Things applications and other environments with limited power or resource consumption. Especially in the edge computing landscape, where computing power and energy consumption are critical factors, the technology offers significant advantages. With this research, the DLR Institute for AI Safety and Security is making an important contribution to the development of sustainable and efficient AI technologies. The presentation at the internationally renowned FPGA Developers' Forum at CERN emphasises the scientific relevance and potential of the development. The full presentation is available on the CERN Document Server and on YouTube and offers interested experts detailed insights into the technical aspects of the innovation.
