The bird population is rapidly declining due to deforestation, agriculture and climate change. Scientists are tracking the types, recording their screams, but even the best computer cannot reliably distinguish the signals of some birds from others. In any case, could not. Now, thanks to the small and large crowdsourcing artificial intelligence scientists cleaned a little feathers this issue.
Artificial intelligence algorithms can be capricious finches, often requiring manual calibration and requalification for each new location or look. Therefore, an interdisciplinary team of researchers launched a Bird Audio Detection program, which got many hours of audio recordings of the observation stations near Chernobyl in Ukraine (they were able to access them), and crowd-sourced entries, some of which came from the app Warblr.
People tagged the 10-second recording or the like containing a bird’s call or not. Using machine learning where computers are trained on the available data — 30 teams teach their AI in the set of records, which were affixed labels and then tested on the records, you do the program or not. Relying in their work on principles of neural networking, artificial intelligence worked like the brain, linking many small computing elements, similar to neurons.
At the end of the monthly competition for the best algorithm scored a 89 out of 100 points on the statistical system performance evaluation AUC. Larger number in this case indicates that the algorithm was able to avoid defining non-avian sounds such as bird (if it was humans, insects, or some rain) and he did not miss the sounds of real birds (even in the case of weak records).
At the moment the algorithms are unable to outperform the people (who actually mark the entries with labels), but the machines can work day and night not afraid of the rain. Was only a matter of time when AI, hatched from this competition, will fly in the real world.