Logo des Repositoriums

Auflistung nach Autor:in "Brust, Clemens-Alexander"

1 - 3 von 3
  • Zeitschriftenartikel
    Active and Incremental Learning with Weak Supervision
    (KI - Künstliche Intelligenz: Vol. 34, No. 2, 2020) Brust, Clemens-Alexander; Käding, Christoph; Denzler, Joachim
    Large amounts of labeled training data are one of the main contributors to the great success that deep models have achieved in the past. Label acquisition for tasks other than benchmarks can pose a challenge due to requirements of both funding and expertise. By selecting unlabeled examples that are promising in terms of model improvement and only asking for respective labels, active learning can increase the efficiency of the labeling process in terms of time and cost. In this work, we describe combinations of an incremental learning scheme and methods of active learning. These allow for continuous exploration of newly observed unlabeled data. We describe selection criteria based on model uncertainty as well as expected model output change (EMOC). An object detection task is evaluated in a continuous exploration context on the PASCAL VOC dataset. We also validate a weakly supervised system based on active and incremental learning in a real-world biodiversity application where images from camera traps are analyzed. Labeling only 32 images by accepting or rejecting proposals generated by our method yields an increase in accuracy from 25.4 to 42.6%.
  • Textdokument
    Carpe Diem: A Lifelong Learning Tool for Automated Wildlife Surveillance
    (INFORMATIK 2021, 2021) Brust, Clemens-Alexander; Barz, Björn; Denzler, Joachim
    We introduce Carpe Diem, an interactive tool for object detection tasks such as automated wildlife surveillance. It reduces the annotation effort by a utomatically selecting informative images for annotation, facilitates the annotation process by proposing likely objects and labels, and accelerates the integration of new labels into the deep neural network model by avoiding re-training from scratch. Carpe Diem implements active learning, which intelligently explores unlabeled data and only selects valuable examples to avoid redundant annotations. This strategy saves expensive human resources. Moreover, incremental learning enables a continually improving model. Whenever new annotations are available, the model can be updated efficiently and quickly, without re-training, and regardless of the amount of accumulated training data. Because there is no single large training step, the model can be used to make predictions at any time. We exploit this in our annotation process, where users only confirm or reject proposals instead of manually drawing bounding boxes.
  • Konferenzbeitrag
    Towards Enabling Level 3A AI in Avionic Platforms
    (Software Engineering 2023 Workshops, 2023) Zaeske, Wanja; Brust, Clemens-Alexander; Lund, Andreas; Durak, Umut
    The role of AI evolves from human assistance over human/machine collaboration towards fully autonomous systems. As the push towards more autonomy subsequently removes the reliance on a human overseeing the system, means of self supervision must be provided to enable safe operations. This work explores dynamic reconfiguration to provide resilience to unforeseen environmental conditions that exceed the systems capabilities, but also against normal faults. We focus on providing the means for this in an ARINC 653 compliant environment, since we target avionics platforms. Scheduling and communication are two major aspects of dynamic reconfiguration. Hence, we discuss multiple respective implementation approaches. The third pillar of reconfiguration, the process of deciding when to reconfigure is also investigated. Combining these yields the building blocks for a self-supervising system.