Researchers have competed fiercely for years to build computational models of the brain with an ever-larger number of simulated neurons, but scientists from the University of Waterloo have taken a different approach: they have built a model to explain how brain activity generates complex behavior.
In a paper published today in Science, Chris Eliasmith and his colleagues describe "Spaun," a 2.5-million-neuron model of the brain they hope will help bridge the brain-behavior gap. Spaun, short for Semantic Pointer Architecture Unified Network, can recognize numbers, generate answers to simple numerical questions, and write them down using a physically modeled arm.
In their experiments the researchers presented Spaun with images of handwritten or typed characters, mostly numbers. Spaun processes them in various ways, and writes responses. For example, if Spaun is presented the number six, it can identify it and recreate it.
The incoming visual image is first compressed to extract the essential visual elements. Spaun's working memory system is composed of a high-dimensional neural integrator, taken from computational neuroscience, and convolution memories taken from mathematical psychology. The neural integrator allows Spaun to store information and the convolution memories provide a memory-efficient algorithm that allows Spaun to bind newly arriving information with a representation indicating its syntactic role. The computations to turn that into arm movements are based on optimal control theory.
Previous brain model projects, such as the Blue Brain Project (1 million neurons), IBM's SyNAPSE Cognitive Computing Project (1 billion neurons, or a bit larger than a cat brain), and a human-scale simulation of 100 billion neurons have been reported. These projects are impressive in scale, but simulating large numbers of neurons alone doesn't explain how complex brain activity generates complex behavior. Eliasmith's 2.5-million spiking neuron model attempts to address the brain-behavior gap, a central challenge in neuroscience.
Spaun can peform eight different tasks involving recognizing numbers and performing motor responses, but the model is hard-wired and cannot learn new tasks. The learning issues are a principle shortcoming, but perhaps a wise one to sidestep at this point, says Christian Machens at the Champalimaud Neuroscience Programme in Lisbon, in a related article also published today in Science. Machens says other than this shortcoming, Eliasmith's model provides a coherent theory on how the brain works. He writes: "To paraphrase the statistician George Box, their model is likely to be wrong, but is certainly useful."
Photo and video: Chris Eliasmith et. al.
Emily Waltz is a features editor at Spectrum covering power and energy. Prior to joining the staff in January 2024, Emily spent 18 years as a freelance journalist covering biotechnology, primarily for the Nature research journals and Spectrum. Her work has also appeared in Scientific American, Discover, Outside, and the New York Times. Emily has a master's degree from Columbia University Graduate School of Journalism and an undergraduate degree from Vanderbilt University. With every word she writes, Emily strives to say something true and useful. She posts on Twitter/X @EmWaltz and her portfolio can be found on her website.