Imagine an aquarium with 100 fish of the same species swimming in it. Then try to track down one of them without getting lost. Tricky, right? What happens if you close your eyes for a few seconds? Then it’s almost impossible. But that was the starting point for a group of researchers at the Institute for Collective Behavior at the Champalimaud Center for the Unknown (CCU) in Lisbon, led by Spanish researcher Gonzalo de Polavieja. team is, AI system would be able to accomplish such a feat. After all, de Polavieja had already tried his luck in his 2014 using conventional algorithms and their results. technology project They barely managed to keep track of dozens of them. The software they are currently developing is based on a deep learning neural network that allows him to track 100 fish with 99.9% accuracy. It is available for download as open source code under the codename Idtracker.ai.
“New AI-based software can track 1 fish out of 100 with 99.9% accuracy.”
The researchers first set up a tank with dozens of zebrafish and a camera on top to track the specimen’s movements. Tests were performed with 30, 50 and 100 fish. It’s a relatively modest number, but it would take traditional software years to identify movement patterns. And, as I pointed out at the beginning of this article, it proves completely infeasible for the human mind. because it increases to
Therefore, when researchers give up using traditional algorithms, Two deep learning neural networks You can clone your brain and learn on your own. The first is used to distinguish fish from other elements in the environment, and the second tracks individual fish by names such as George and Tom. If it remains, it will be cleared by the conventional algorithm. After processing the video feed for 1 hour, we are able to identify 99.9% of the specimens.
Once the process is complete, artificial intelligence software It has already learned who is who in the fish shoal and can recognize George, Tom or any of the other 99 specimens by viewing random snippets of the video. The system is highly scalable and has already been tested with up to 150 fish with very low error margins.
real application
And what interest could there be in identifying zebrafish? The answer lies in collective action. A system based on the Idtracker.ai software could have city security and security. safety applications, to track individuals or study crowd behavior in different situations such as supermarkets and concerts. Similarly, we can establish collaborative processes between people with potential applications in sociology.
sauce: eye sink
