{"id":24031,"date":"2018-05-11T14:41:15","date_gmt":"2018-05-11T18:41:15","guid":{"rendered":"https:\/\/www.tun.com\/blog\/?p=24031"},"modified":"2022-03-16T11:59:18","modified_gmt":"2022-03-16T15:59:18","slug":"columbia-algorithm-animal-behavior","status":"publish","type":"post","link":"https:\/\/www.tun.com\/blog\/columbia-algorithm-animal-behavior\/","title":{"rendered":"AI Can Now Decode Animal Behavior. What Does It Mean for Our Future?"},"content":{"rendered":"

Up until now, biologists studying animal behavior have been restricted to human observation. But scientists at <\/span>Columbia University<\/span><\/a> have now come up with an easier method. They have developed <\/span>an innovative algorithm<\/span><\/a> that can be used to study animal behaviors. <\/span><\/p>\n

Led by <\/span>Rafael Yuste<\/span><\/a>, a neuroscientist at Columbia and a member of <\/span>Columbia\u2019s Data Science Institute<\/span><\/a>, the team designed an algorithm that can successfully analyze hours of video of <\/span>Hydra<\/span><\/a>, a miniscule freshwater invertebrate, and understand its full range of behaviors. <\/span><\/p>\n

By filtering out spam information, the algorithm is capable of detecting tendencies in the animal\u2019s behavior.<\/span><\/p>\n

The full study is published in the journal <\/span>eLife<\/span><\/a>.<\/span><\/p>\n

Animal behavior has been a subject of scientific research since the days of Aristotle, noted <\/span>Shuting Han<\/span><\/a>, a graduate student at Columbia and lead author of the study. However, the process of researching animal behavior is notoriously painstaking, typically consisting of hours of scrupulous observation and note-taking.<\/span><\/p>\n

The new algorithm provides a major shortcut in this kind of research.<\/span><\/p>\n

\u201cBehavioral research has traditionally largely been descriptive and relies on human observation,\u201d said Han. \u201cThese qualitative descriptions are limited by the subjective nature of human annotations, the language we use, and are also very time-consuming. Therefore it is important to develop quantitative and automated methods for behavior analysis, in order to systematically understand the expression and variability of behavior of different animals under different experimental conditions.\u201d<\/span><\/p>\n

The Algorithm<\/b><\/h2>\n

The algorithm developed by the researchers is a \u201cbag-of-words\u201d algorithm, a popular kind of algorithm often used in filtering email spam. Typical bag-of-words algorithms are able to classify the topics of documents and identify ones that could potentially be spams. <\/span><\/p>\n

The researchers applied this same concept in their algorithm, only using videos instead of documents. <\/span><\/p>\n

The algorithm learns to classify different visual patterns — shapes and motions — in videos of Hydra and pick out repetitive movements. <\/span><\/p>\n

In doing so, the algorithm identifies the different behavioral patterns of the animal.<\/span><\/p>\n

The Study<\/b><\/h2>\n

Hydra, a primitive animal whose ancestors appeared on earth around 700 million years ago, are simple organisms with no backbone or brain. It does have a nervous system, however. Hydra has a translucent body with hundreds of neurons that coordinate its movements and behavior.<\/span><\/p>\n

\u201cHydra \u2026 provides a unique opportunity to study the evolution of behaviors and their underlying neural codes,\u201d said Han. \u201cHydra represents one of the most primitive animals with a nervous system, and with modern imaging techniques we can record the complete neural activity pattern during its behaviors.\u201d<\/span><\/p>\n

The researchers can therefore directly compare the activity of Hydra\u2019s neurons to its behavioral activity, allowing them to analyze how its nervous system functions.<\/span><\/p>\n

In a <\/span>previous study<\/span><\/a>, the team recorded the entirety of the animal\u2019s neural activity, and were able to connect four sets of neural circuits to four unique movement patterns, illustrating how its nervous system functions. <\/span><\/p>\n

The new study expanded on this previous research to include not just four, but all of its behavioral patterns. The algorithm was able to detect 10 behaviors and compared six of those behaviors to its environmental conditions.<\/span><\/p>\n