Fraunhofer IIS and Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) bring together research and education to further enhance machine learning
Fraunhofer IIS and FAU are teaming up to dive deeper into research in the field of machine learning. Their recently launched project “KISS” (German abbreviation that corresponds to “Laboratory for system level design of machine learning-based signal processing applications”) investigates new development tools to improve AI-based algorithms in signal processing. The German Federal Ministry of Education and Research approved the research proposal submitted by Fraunhofer and FAU.
Machine learning is being adopted in an increasing number of disciplines. Signal processing can also greatly benefit from machine learning, as AI-based algorithms often produce better results than ordinary techniques. KISS researches novel tools to enhance these algorithms and implement them efficiently on various systems.
KISS elaborates concepts to generate training data for artificial intelligence. There are several methods to this end, including simulating data, creating data sets oneself or accessing existing databases from various sources such as video hosting services. However, these datasets often require large amounts of memory. The project aims to solve these and other technical challenges by enabling automatic translation of a given network description to facilitate efficient implementation on different target architectures such as GPUs or FPGAs.
To ensure practical relevance, the methods developed are tested with various video, audio and speech applications. Relevant concepts include the processing of single- and multichannel microphone signals for human-machine interfaces (HMIs) or an enhanced source separation for hearing aids. With regard to video, the focus will be on algorithms for digitizing environments. Neural networks can generate better depth maps or estimate surface characteristics and lighting situations. These algorithms are used in the film and media industry and in advanced driver-assistance systems.
KISS brings together research and education. Students have the opportunity to write their theses or do an internship in the context of the project. Furthermore, the insights gained during the project will be edited for use in various lectures at FAU. This ensures that there will be an adequate supply of machine learning experts in Germany in the future. In addition, there will be training courses for employees in the industry and scientific publications on the topics. This exchange with the public enables the scientists to learn which problems the industry faces today, so that these issues can be adequately represented in their research.
For more information on the project, please visit our related website.
Header image © fotomek – fotolia.de/Fraunhofer IIS
This post is also available in: Deutsch