Artificial networks teach you to smell like the brain | MIT News
Through machine learning, a computer model can learn to smell itself in just a few minutes. When it does, the researchers found that it builds a neural network that closely mimics the olfactory circuits that animal brains use to process odors.
Animals, from fruit flies to humans, all use essentially the same strategy to process scent information in the brain. But the neuroscientists who trained an artificial neural network to perform a simple task of classifying odors were surprised to see it reproduce biology’s strategy so faithfully.
“The algorithm we are using bears no resemblance to the actual process of evolution,” says Guangyu Robert Yang, associate researcher at the McGovern Institute for Brain Research at MIT, who led the work as a postdoctoral fellow at the University. Columbia. The similarities between artificial and biological systems suggest that the brain’s olfactory network is well suited for its task.
Yang and colleagues, who reported on their findings on October 6 in the newspaper Neuron, say their artificial network will help researchers learn more about the olfactory circuits of the brain. The work is also helping to demonstrate the relevance of artificial neural networks for neuroscience. “By showing that we can match the architecture [of the biological system] very specifically, I think it gives more confidence that these neural networks can continue to be useful tools for modeling the brain, âsays Yang, who is also an assistant professor in the departments of brain and cognitive science and electrical engineering and computer science from MIT.
Map the natural olfactory circuits
For fruit flies, the organism in which the brain’s olfactory circuits have been best mapped, the smell begins in the antennae. Sensory neurons, each equipped with specialized odor receptors to detect specific odors, transform the binding of odor molecules into electrical activity. When an odor is detected, these neurons, which make up the first layer of the olfactory network, send a signal to the second layer: a collection of neurons that reside in a part of the brain called the antennal lobe. In the antennal lobe, sensory neurons that share the same receptor converge on the same second-layer neuron. âThey are very demanding,â says Yang. “They don’t get any input from neurons expressing other receptors.” Because it has fewer neurons than the first layer, this part of the network is considered a compression layer. These second-layer neurons, in turn, send signals to more third-layer neurons. Oddly enough, these connections appear to be random.
For Yang, a computational neuroscientist, and Peter Yiliu Wang, a graduate student at Columbia University, this knowledge of the olfactory system of the fly presented a unique opportunity. Few parts of the brain have been mapped so thoroughly, making it difficult to assess the quality of some computer models representing the true architecture of neural circuits, they say.
Build a network of artificial odors
Neural networks, in which artificial neurons rewire themselves to perform specific tasks, are computer tools inspired by the brain. They can be trained to spot patterns in complex data sets, making them valuable for speech and image recognition and other forms of artificial intelligence. There are clues that the neural networks that do this best mimic the activity of the nervous system. But, says Wang, who is now a post-doctoral fellow at Stanford University, differently structured networks could generate similar results, and neuroscientists have yet to know whether artificial neural networks reflect the actual structure of biological circuits. With comprehensive anatomical data on the olfactory circuits of the fruit fly, he says, âwe are able to ask this question: Can artificial neural networks really be used to study the brain? “
Working closely with Columbia neuroscientists Richard Axel and Larry Abbott, Yang and Wang built an artificial neural network comprising an input layer, a compression layer and an expansion layer, much like the olfactory system of the fly. fruits. They gave it the same number of neurons as the fruit fly system, but no inherent structure: the connections between neurons would be rewired as the model learned to classify smells.
Scientists asked the network to assign data representing different smells to categories and correctly categorize not only individual smells, but also mixtures of odors. This is something the brain’s olfactory system is particularly good at, Yang says. If you combine the scents of two different apples, he explains, the brain still smells like an apple. On the other hand, if two photographs of cats are mixed pixel by pixel, the brain no longer sees a cat. This ability is only a hallmark of the brain’s odor-processing circuitry, but captures the essence of the system, Yang says.
It only took a few minutes for the artificial network to organize itself. The structure that emerged was surprisingly similar to that found in the brain of the fruit fly. Each neuron in the compression layer received inputs from a particular type of input neuron and connected, apparently randomly, to several neurons in the expansion layer. In addition, each neuron in the expansion layer receives connections, on average, from six neurons in the compression layer – just as happens in the brain of the fruit fly.
âIt could have been one, it could have been 50. It could have been anywhere in between,â Yang said. “Biology finds six, and our network finds about six as well.” Evolution has found this organization by random mutation and natural selection; the artificial network found it through standard machine learning algorithms.
The surprising convergence proves that the brain circuits that interpret olfactory information are optimally organized for their task, he says. Now, researchers can use the model to further explore this structure, exploring how the network evolves under different conditions and manipulating the circuits in ways that cannot be achieved experimentally.