1 - 10 of 14
-
PublicationKI-Engineering in der Produktion(Fraunhofer Verlag, 2023)
-
PublicationInformed Machine Learning - A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems( 2023)Despite its great success, machine learning can have its limits when dealing with insufficient training data. A potential solution is the additional integration of prior knowledge into the training process which leads to the notion of informed machine learning. In this paper, we present a structured overview of various approaches in this field. We provide a definition and propose a concept for informed machine learning which illustrates its building blocks and distinguishes it from conventional machine learning. We introduce a taxonomy that serves as a classification framework for informed machine learning approaches. It considers the source of knowledge, its representation, and its integration into the machine learning pipeline. Based on this taxonomy, we survey related research and describe how different knowledge representations such as algebraic equations, logic rules, or simulation results can be used in learning systems. This evaluation of numerous papers on the basis of our taxonomy uncovers key methods in the field of informed machine learning.
-
PublicationDeutsche Normungsroadmap Künstliche Intelligenz(DIN, 2022-12-09)
-
PublicationNew active learning algorithms for near-infrared spectroscopy in agricultural applications( 2021)The selection of training data determines the quality of a chemometric calibration model. In order to cover the entire parameter space of known influencing parameters, an experimental design is usually created. Nevertheless, even with a carefully prepared Design of Experiment (DoE), redundant reference analyses are often performed during the analysis of agricultural products. Because the number of possible reference analyses is usually very limited, the presented active learning approaches are intended to provide a tool for better selection of training samples.
-
PublicationGenerative Machine Learning for Resource-Aware 5G and IoT SystemsExtrapolations predict that the sheer number of Internet-of-Things (IoT) devices will exceed 40 billion in the next five years. Hand-crafting specialized energy models and monitoring sub-systems for each type of device is error prone, costly, and sometimes infeasible. In order to detect abnormal or faulty behavior as well as inefficient resource usage autonomously, it is of tremendous importance to endow upcoming IoT and 5G devices with sufficient intelligence to deduce an energy model from their own resource usage data. Such models can in-turn be applied to predict upcoming resource consumption and to detect system behavior that deviates from normal states. To this end, we investigate a special class of undirected probabilistic graphical model, the so-called integer Markov random fields (IntMRF). On the one hand, this model learns a full generative probability distribution over all possible states of the system-allowing us to predict system states and to measure the probability of observed states. On the other hand, IntMRFs are themselves designed to consume as less resources as possible-e.g., faithful modelling of systems with an exponentially large number of states, by using only 8-bit unsigned integer arithmetic and less than 16KB memory. We explain how IntMRFs can be applied to model the resource consumption and the system behavior of an IoT device and a 5G core network component, both under various workloads. Our results suggest, that the machine learning model can represent important characteristics of our two test systems and deliver reasonable predictions of the power consumption.
-
-
PublicationIDS Reference Architecture Model. Industrial Data Space. Version 2.0(International Data Spaces Association, 2018)
-
PublicationReference Architecture Model for the Industrial Data Space(Fraunhofer-Gesellschaft, 2017)
-
PublicationInnovative technologies as enabler for sorting of black plastics( 2016)Thermal recycling of plastics is no longer seasonable. More modern recycling techniques require pure fractions containing only a single variety of polymer. A large portion of the plastic waste contains black or multilayer materials that are not sortable with todays sorting technologies. A number of means to analyse and sort mixed plastic waste based on the specific mechanical, electrical, and chemical properties of its components such as density, conductivity and melting point have been developed. The most promising electromagnetic principles like XRay imaging employs ionizing radiation that requires special safety measures, while infrared and visible light is absorbed by the carbon in black plastics. Publications in the last few years show the feasibility of identifying and then separating different types of plastics based on their specific frequency response in the millimetre-wave and terahertz region. From an economical point of view, a line camera radar system operating between 30 GHz and 300 GHz offers an acceptable trade-off between cost, resolution and enough information to reliably identify different materials. The system approach uses data-driven statistical machine-learning methods for classification. The use of deep neural networks in combination with very large training-datasets with thousands of samples improves the predicted sorting purity between 90% and 99.9% for common use-cases. Finally the THz-camera and the classification methods have to be integrated in a sorting solution that meets the realtime requirements of recycling systems. Due to the modular app roach, it is also possible to upgrade existing sorting systems.
-
PublicationSorting of black plastics using statistical pattern recognition on terahertz frequency domain data( 2016)The sorting of used plastics is an ever-growing market field which is further pushed by new EU regulations in, e.g., car recycling. Modern recycling techniques require pure or almost pure fractions of polymers. These pure fractions can be generated from waste using modern sorting technologies based on specific mechanical, electrical and chemical material properties such as density, conductivity and melting point. The thermal recycling of plastics is no longer seasonable. More modern recycling techniques require pure fractions containing only a single variety of polymer. A large portion of the plastic waste contains black or multilayer materials that are not sortable with todays' sorting technologies. To overcome this challenge, three Fraunhofer institutes are working together to develop a new type of sorting system. As a first step, we have developed a frequency domain line-scan camera working in the terahertz range with frequencies below 300 GHz. Since the entropy in terahertz signals below 300 GHz is not as high as needed for simple classification, more complex statistical pattern recognition methods are needed. The application of those methods to the problem of sorting black plastics as the second step in this joint project is presented in this paper. These methods have to be integrated into a real sorting system, which is the third part of our joint project. The modular approach gives the ability to integrate our sensors and algorithms into existing sorting systems.
