The need for training data for machine learning algorithms is still incredibly high and is constantly growing with the growing use of artificial intelligence. Many companies are therefore faced with the challenge of reluctant to use methods such as deep learning because training the models would be too complex and time-consuming. At the same time, however, models available on the market are usually not directly applicable and do not fit optimally to one’s own needs.
In this lecture I would like to report on our experiences with the acquisition of training data in the field of audio and video mining and the corresponding adaptation of our machine learning methods. Especially our ideas and experiences on the applicability of new concepts like Transfer Learning or Active Learning will be discussed.