Artificial intelligence (AI)–powered neural networks modeled on real human brain connectivity patterns perform cognitive tasks better than traditional AI systems, new research suggests.
“This work opens new opportunities to discover how the network organization of the brain optimizes cognitive capacity,” write the researchers from The Neuro (Montreal Neurological Institute-Hospital) and the Quebec Artificial Intelligence Institute, Quebec, Canada.
Senior investigator Bratislav Misic, PhD, said the research has potential clinical application for studying diseases of the brain, which is something his team is actively working on.
“For example, using MRI techniques, we can measure different patterns of atrophy in neurodegenerative diseases such as Alzheimer’s disease,” Misic told Medscape Medical News.
“We can use these disease patterns from real patients to artificially lesion these connectomes and to ask how a particular disease causes a particular pattern of symptoms and cognitive deficits,” he added.
The findings were published online August 9 in Nature Machine Intelligence.
Using brain imaging data, the investigators reconstructed a human brain connectivity pattern and applied it to an artificial neural network (ANN).
After training, the ANN successfully performed a working memory task more flexibly and efficiently than other “benchmark” AI systems.
The researchers note that their approach is unique because previous work on brain connectivity, also known as connectomics, has focused on describing brain organization without regard to how it actually functions.
Traditional ANNs have arbitrary structures that do not reflect how real brain networks are organized. Integrating brain connectomics into the construction of ANNs can reveal how the wiring of the brain supports specific cognitive skills, the investigators write.
“Up until now, if you look at how neural networks are constructed, the architectures that are used are very ad hoc and very problem specific,” Misic said.
“But the connectomics revolution that’s happened in neuroscience over the past 20 years or so has given us the ability to really measure and trace out connection patterns in a variety of organisms, including the human brain,” Misic added.
He noted that the researchers took wiring patterns of the real human brain and implemented it as an ANN. They then “trained that network to perform a very simple cognitive task, and when you compare it to other benchmark architectures, it actually does better,” he said.
This shows that there is “something fundamentally different about how the human brain is wired up and that the design principles that we can see in the human brain could be used to potentially build better artificial networks,” Misic concluded.
Funding for the research was provided by the Canada First Research Excellence Fund, awarded to McGill University for the Healthy Brains, Healthy Lives initiative, and by the Natural Sciences and Engineering Research Council of Canada, Fonds de Recherche du Quebec – Santé, the Canadian Institute for Advanced Research, Canada Research Chairs, Fonds de Recherche du Quebec – Nature et Technologies, and the Centre UNIQUE (Union of Neuroscience and Artificial Intelligence). The investigators have reported no relevant financial relationships.
Nat Mach Intell. Published online August 9, 2021. Abstract