Christian Witte

"Active and Continuous Learning for Autonomous Driving"

Incorporating unseen data in pre-trained neural networks remains a challenging endeavor, as a complete retraining is often impracticable. Yet, training the networks sequentially on data with different distributions can lead to a performance degradation for previous learned data, known as catastrophic forgetting. The sequential training paradigm and the mitigation of catastrophic forgetting is subject to Continual Learning (CL). The phenomenon of forgetting poses a challenge for applications with changing distributions and prediction objectives, including Autonomous Driving (AD). Further, modern computer vision requires extensive amount of labeled data for training. Annotating the data with ground truth information is an expensive and cumbersome manual process. To reduce this effort, Active Learning (AL) considers the selection of subset(s) for the annotation process, such that networks trained on these subsets attain a similar performance if trained on the entire dataset.

Termin als Datei downloaden und in den Kalender importieren.