What does Deep Learning have to do with animal behaviors?

https://www.youtube.com/watch?v=lVNSW0hQrlw&feature=youtu.be

To study animal behaviors, different methods involving Deep Learning algorithms have been developed. Let’s have a look at one of the most outstanding project in the field.

If you do believe (as I did before approaching this topic) that capturing the posture of animals has to do with selfie poses, you are far from reality. Indeed, pose estimation concerns the study of the geometrical configuration of animal body parts. This knowledge leads to a better understanding of animal behaviors and animal brains. Therefore, behavioral studies are not only used to set biodiversity preservation plans but play also a key role in the area of neuroscience.

To be precise, humans are documenting animal postures since the first cave paintings, in the Old Stone Age. However, studying behaviors is extremely complex and can not merely rely on human observations. Nowadays, cave art is not an option anymore and ecologists and conservationists had to adapt. Sensitive cameras are therefore used to monitor protected areas, as the one implemented more than ten years ago in the Serengeti National Park and Grumeti Reserve in Tanzania. The cameras capture pictures or videos of the animals living in the area, thus allowing researchers to track and study animal movements, behaviors, and demographics. In the example mentioned, the photos are collected and stored on a web-based platform called Zooniverse. Afterward, researchers from around the world are working very intensely together to identify animals, species, and to label the picture accordingly.  However, even if some improvements have been made compared to the Paleolithic time, you can certainly imagine how slow and time-consuming this process is.

Deep Learning has extremely accelerated this procedure. Scientists have built up a Deep Learning toolkit called DeepPoseKit which measures the position of animal body parts in the space by using data (videos or pictures). This tool combines previous estimations techniques with state-of-the-art technology. It can quickly and accurately track animal postures, after being trained with just 100 examples. (If you don’t understand what this last sentence means, check my blog posts about how Deep Learning works, and how it is trained.) DeepPoseKit has solved the trade-off speed-accuracy that was present in the previous approaches and moreover, the tool can be applied across species and experimental conditions – from hornets to hippopotamus, from laboratories to wild scenarios. Through the data collected, it is possible to study animals, their interactions within the group, and the surroundings. This algorithm is trained to automatically identify, label, and count animals by reducing to almost zero human intervention. This means, starting from a training set of data, Deep Learning algorithms can automatically estimate the location of animal body parts. After some training, the tool can generalize the findings and apply the elaborated rules to a new set of data.

DeepPoseKit can be beneficial also for experts in the field of psychology, neuroscience, and ecology. Indeed, this algorithm can be used to monitor and study human movement as well, such as movement-related diseases.

It seems that Deep Learning applications are endless and omnipresent, and they can be implemented in almost every field. Therefore, for the next blog post, I’d like you to decide about the topic! 🙂
At this link, you find a poll where you can insert the topic of Deep Learning that fascinates you the most / the sector in which you think Deep Learning applications could make a difference. You can also directly scan this QR code to go to the poll:

I’ll pick up the most interesting and write the next article about it! 🙂

References:

Photo by Sam Carter on Unsplash

Using machine learning to accelerate ecological research

Machine learning animal poses to understand the behavior

DeepPoseKit, a software toolkit for fast and robust animal pose estimation using deep learning

Deep learning tools for the measurement of animal behavior in neuroscience

New method improves measurement of animal behavior using deep learning