Logo des Repositoriums

Auflistung nach Schlagwort "Transfer Learning"

1 - 5 von 5
  • Konferenzbeitrag
    A Children's Toy for Learning AI
    (Mensch und Computer 2019 - Tagungsband, 2019) Scheidt, Alexander; Pulver, Tim
    Here we present Any-Cubes, a prototype toy with which children can intuitively and playfully explore and understand machine learning as well as Internet of Things technology. Our prototype is a combination of deep learning-based image classification [9] and machine-to-machine (m2m) communication via MQTT. The system consists of two physical and tangible wooden cubes. Cube 1 ("sensor cube") is inspired by Google’s teachable machine [11,12]. The sensor cube can be trained on any object or scenery. The machine learning functionality is directly implemented on the microcontroller (Raspberry Pi) by a Google Edge TPU Stick. Via MQTT protocol, the microcontroller sends its current status to Cube 2, the actuator cube. The actuator cube provides three switches (relays controlled by an Arduino board) to which peripheral devices can be connected. This allows simple if-then functions to be executed in real time, regardless of location. We envision our system as an intuitive didactic tool for schools and maker spaces.
  • Textdokument
    Exploring Texture Transfer Learning via Convolutional Neural Networks for Iris Super Resolution
    (BIOSIG 2017, 2017) Ribeiro,Eduardo; Uhl,Andreas
    Increasingly, iris recognition towards more relaxed conditions has issued a new superresolution field direction. In this work we evaluate the use of deep learning and transfer learning for single image super resolution applied to iris recognition. For this purpose, we explore if the nature of the images as well as if the pattern from the iris can influence the CNN transfer learning and, consequently, the results in the recognition process. The good results obtained by the texture transfer learning using a deep architecture suggest that features learned by Convolutional Neural Networks used for image super-resolution can be highly relevant to increase iris recognition rate.
  • Konferenzbeitrag
    Knowledge Self-Adaptive Multi-Agent Learning
    (INFORMATIK 2019: 50 Jahre Gesellschaft für Informatik – Informatik für Gesellschaft (Workshop-Beiträge), 2019) Reichhuber, Simon
    In this paper concepts of a starting Doctoral Dissertation are presented, discussing the question how agents constructed according to Organic Computing methodologies can autonomously identify Knowledge Sources and adapt them to their learning procedure. Achieving this, the fields of Multi-Agent Learning, Organic Computing, Transfer Learning, and Online Learning are combined to an unified architecture. The focus of the work is on the real-time evaluation of knowledge sources. In order to show the practical use case of such systems, the author presents two scenarios. The first, collaborative crawling, is an information retrieval task, hence it deals with knowledge distributed over multiple websites. Whereas the latter is designed to run in a virtual space, the second, denoted as machine park collaboration, can be implemented in industrial 4.0 fields of the real world.
  • Konferenzbeitrag
    Landwirtschaftliche Ertragsvorhersage im Kontext begrenzter realer Trainingsdatensätze: ein Transfer-Learning-Ansatz unter Verwendung tieferneuronalerNetze
    (INFORMATIK 2023 - Designing Futures: Zukünfte gestalten, 2023) Münzberg, Alexander; Troost, Christian; Reinosch, Nils; Martini, Daniel; Seuring, Liv; Niehus, Alexander; Srivastava, Rajiv; Streck, Thilo; Berger, Thomas; Bernardi, Ansgar
    Anhand von KI-gestützten Entscheidungshilfen in der Landwirtschaft, beispielsweise durch Anpassung von Düngeapplikationen oder des zeitlichen Feldarbeits-Managements, kann die Produktivität auf einer ökologischen und nachhaltigen Sicht gesteigert werden. Wir beschreiben eine Lösung, um mit Hilfe von neuronalen Netzen Ertrags- und Wachstumsprognosen in realen landwirtschaftlichen Daten zu erzielen. Das Problem geringer Trainingsdatenmengen wird dadurch gelöst, dass zunächst ein System anhand von Simulationsdaten antrainiert und mittels Transfer- Learning an spezifische reale Betriebsbedingungen anhand einiger weniger Realdaten angepasst wird. Die Ergebnisse der Realprognose werden anhand einer Kreuzvalidierungsstrategie evaluiert.
  • Textdokument
    Using Transfer Learning for Quality Improved Forecasting of Temporal Agricultural Processes by Adapting Convolutional Neural Networks
    (INFORMATIK 2022, 2022) Münzberg,Alexander; Troost,Christian; Bernardi,Ansgar
    AI-based decision support can help farmers to reach improved productivity in an environmentally sustainable way. Through transfer learning, an existing Convolutional Neural Network is progressively adapted to provide high quality forecasting results using agricultural time series in the context of different locations, growth and soil types, climate zones, and management variations. The delivered results are validated by appropriate statistical methods and show improved prediction accuracy.