Though we might assume that a fruit fly follows its nose to the scent of that sweet ripening fruit on the counter, their vision might also be a guiding force. And, even more surprisingly, they could actually be flitting over because they see another fly that they recognize.
In order to investigate what fruit flies, or Drosophila melanogaster in scientific speak, can theoretically see and understand, biology researchers at UTM have developed a machine-learning model to better understand the fly’s visual system and ability for visual comprehension. Their findings, which have so far concluded that visual learning in fruit flies has much higher capacity than previously believed, have been published in the October edition (Vol.13, Issue 10) of PLOS ONE.
· PLOS ONE article
“This machine-learning model can identify other flies as individuals with unique and distinctive visual characteristics,” says senior author Professor Joel Levine, Chair of UTM’s Department of Biology, Canadian Institute for Advanced Research (CIFAR) Senior Fellow, and Canada Research Chair in Mechanisms and Features of Social Behaviour.
“In our study we found that the flies were able to recognize individual Drosophila with a remarkable degree of accuracy, something that even humans have trouble with,” says Levine. He emphasizes the unique interdisciplinarity of the project that involved artificial intelligence (AI) and biology, as well as deep learning, a budding subfield connected to machine learning.
This study is a collaboration with colleagues in the School of Engineering at the University of Guelph under the direction of Graham Taylor, Canada Research Chair in Machine Learning. Taylor is a collaborator on a number of initiatives in Ottawa and Toronto and is a prominent scholar in Canada’s AI community, as well as a CIFAR Azrieli Global Scholar.
The study’s lead author, CIFAR postdoctoral fellow Jonathan Schneider marvels at the seeming simplicity of the fruit fly, which are surprisingly far more complex and fascinating than they at first appear.
“The fact that the fly eye’s anatomy is wired like a convolutional network is just one of the reasons Drosophila is such a great organism to bridge machine learning and biology,” says Schneider.
“By leveraging maps of the brain to build a deep network model wired like the fly, we were able to show Drosophila’s surprising ability to recognize details and to know each other individually. Models like these can help us probe into how the fly, and all of us, process and understand the visual world.”