Systematische merkmalsbewertung in komplexen ultraschallsignalen mit lernmaschinen

Bats are capable of taking very complex decisions based on ultrasound signals. So far, it remained elusive which features allow for the astonishing behavioural abilities of these animals. We show that machine learning methodology leads to novel and sometimes surprising insights into this question using two examples: the determination of the plant species fromits ultrasound echo and the recognition of a conspecific bat individuum from its echolocation call. The learning machines are used to systematically examine a large number of candidate features and evaluate them in terms of their relevance for the discrimination task.