Reading the minds of animals with artificial intelligence
Animals as well as humans have emotions.
Many people may feel that they can somehow understand the joy, anger, sorrow, and pleasure of the dogs or cats they keep at home.
However, evaluating the emotions of small laboratory animals, such as mice, is a very difficult problem.
At present, the evaluation of emotions in laboratory animals is being carried out in various fields, from basic research to elucidate brain function to applied research for the development of new drugs.
However, many of the current methods involve humans observing animal behavior and assigning scores, which inevitably introduces the observer’s subjectivity.
I am developing methods to evaluate animal behavior objectively, using artificial intelligence techniques to reduce human bias.
If we accurately evaluate animal emotions, we can improve the relationship between animals and humans.
In addition, I am studying immune responses in cancers and how food affects our health.
Educational approach
Integrating machine learning and animal behavior analysis
In recent years, with the development of computer science, it has become possible to efficiently process and analyze very large datasets, known as big data.
In the field of biology as well, it is common to measure many genes and proteins simultaneously and to analyze the vast amounts of data to test hypotheses.
In other words, in addition to traditional methods such as culturing cells with pipettes or observing laboratory animals, mathematical and computer methods are increasingly essential.
I am conducting research that uses machine learning (artificial intelligence) techniques to automatically analyze animal behavior from videos recording the behavior of laboratory animals.
The field is still young and evolving, which makes it an exciting place to learn by doing. If this sparks your interest, consider joining us and becoming a“biologist who can code.”
Vision for industry-academia collaboration
Creating objective systems for evaluating animal behavior with machine learning
Behavioral observation of animals is the most basic method for understanding their condition, and it is conducted in many fields, both basic and applied research.
However, current methods of behavioral observation rely on human observation, and thus suffer from low objectivity, poor reproducibility, and limited throughput.
By using machine learning methods such as neural networks, I am developing animal behavior analysis methods that are automatic, objective, and high-throughput.
Previously, I reported a method to accurately distinguish scratching and grooming behaviors in mice from footage recorded with a handy camera (Scientific Reports 2021, Frontiers in Behavioral Neuroscience 2022).
In the future, I plan to analyze other behaviors such as exploratory activity, drinking, and feeding, as well as to use the developed methods to analyze disease models and explore novel phenotypes.
