Using Deep Learning to make Better Models of the Brain
ERC Consolidator Grant for Tübingen AI researcher Professor Jakob Macke
Tübingen AI researcher, Professor Jakob Macke, is aiming to use Deep Learning methods to build new neuronal networks at the intersection of neuroscience and machine learning. The project ‘DeepCoMechTome’, which is receiving close to two million euros from the European Research Council, is focused on the brain of the fruit fly Drosophila melanogaster. The fruit fly’s brain has over 100,000 nerve cells and several million connections.
The brain of Drosophila, which is one of the best known model organisms in biology, is interesting to both neuroscientists and computer scientists for several reasons. “Fruit flies only have tiny brains in comparison to mammals,” explains Macke, “Nevertheless, these animals have astonishing capabilities, such as highly precise control of their flight. They can use landmarks to orient themselves and respond rapidly when predators approach.” Despite its massive capacity, the brain of Drosophila only uses nanowatts of energy, making it far more energy-efficient than any computer.
In recent decades biological research has succeeded in analyzing the brain of the fruit fly at a high level of detail. “We now have access to the brain’s circuitry, but until now we haven’t been able to build any models that can solve tasks which are as challenging as those solved by real Drosophila brains.” Meanwhile, recent advances in Deep Learning have made it possible to create artificial neuronal networks that can carry out extremely complicated calculations. However, these artificial networks have become very different in architecture and functioning to biological ones.
“We want to bring these two worlds back together,” explains the AI researcher, “Our goal now is to create artificial neuronal networks that, in composition and structure, are similar to the brain of fruit flies, but at the same time can perform similarly complex computations.” This requires innovative machine learning methods, which will be developed and applied in the project, in close cooperation with Srinivas Turaga from the Janelia Research Campus in Ashburn, Virginia, and other researchers. Such methods could also be used for neuroscientific research into animals with more complex brains, such as fish or mammals.
Dr. Jakob Macke (born 1982) has been appointed to the professorship for Machine Learning in Science at the Faculty of Mathematics and Natural Sciences at the University of Tübingen. This professorship was established in the Cluster of Excellence “Machine Learning – New Perspectives for the Sciences” as of the summer semester 2020. Macke is also a member of the Tübingen AI Center, which is funded by the federal government, and is speaker of the “Bernstein Center for Computational Neuroscience Tübingen”.