New Technique Leads to more Brainlike AI

Researchers at the Dutch National Research Institute for Mathematics and Computer Science (CWI), a member of the Human Brain Project (HBP), have developed a new technique that could allow an artificial intelligence to behave more like a human brain. The study appears in the journal Nature Machine Intelligence based on the work of CWI’s Bojian Yin and Sander Bohke.

Spiking neural networks are a special type of neural network that more closely resembles the brain, however, the interactions between the nodes in a spiking neural network require a lot of memory, which limits the size. The researchers hoped that, by making the AI learn the way a human learns, they could reduce the complexity and therefore scale to a larger size.

According to researcher Bojian Yin, “The learning aspect of these algorithms is a big challenge, and they cannot match the learning ability of our brain. The brain can easily learn immediately from new experiences, by changing connections, or even by making new ones. The brain also needs far fewer examples to learn something and it works more energy-efficiently. We wanted to develop something closer to the way our brain learns… If you make a mistake during a driving lesson, you learn from it immediately. You correct your behavior right away and not an hour later. "You learn, as it were, while taking in the new information. We wanted to mimic that by giving each neuron of the neural network a bit of information that is constantly updated. That way, the network learns how the information changes and doesn't have to remember all the previous information.”

Making it easier to learn from the data allows larger spiking neural networks. Sander Bohte adds, "Previously, we could train neural networks with up to 10,000 neurons; now, we can do the same quite easily for networks with more than six million neurons."

For more information, see the press release at the Human Brain Project.