Brain Cell Differences Could Be Key to Learning in Humans and AI | Imperial News

Imperial researchers have found that variability between brain cells could speed up learning and improve brain performance and future AI.

The new study found that by adjusting the electrical properties of individual cells in brain network simulations, the networks learned faster than simulations with identical cells.

Having a diversity of neurons in the brain and AI … could stimulate learning. Nicolas perez Department of Electrical and Electronic Engineering

They also found that the arrays required fewer modified cells to achieve the same results and that the method consumes less power than models with identical cells.

The authors say their findings could teach us why our brains are so good at learning, and could also help us build better artificially intelligent systems, such as digital assistants that can recognize voices and faces, or technology. self-driving cars.

First author Nicolas perez, doctoral student at Imperial College London Department of Electrical and Electronic Engineering, said: “The brain must be energy efficient while being able to excel at solving complex tasks. Our work suggests that having a diversity of neurons in the brain and in AI meets both of these requirements and may stimulate learning. ”

The research is published in Nature Communication.

Why does a neuron look like a snowflake?

Our research suggests that we can learn vital lessons from our own biology to make AI work better for us. Dr Dan Goodman Department of Electrical and Electronic Engineering

The brain is made up of billions of cells called neurons, which are connected by vast “neural networks” that allow us to learn more about the world. Neurons are like snowflakes: they look alike from afar, but on closer inspection, it’s clear that no two are exactly alike.

In contrast, every cell in an artificial neural network – the technology behind AI – is identical, only their connectivity varying. Despite the speed at which AI technology is advancing, their neural networks don’t learn as precisely or quickly as the human brain – and researchers wondered if their lack of cellular variability could be a culprit.

They set out to study whether brain emulation by varying the properties of neural network cells could stimulate AI learning. They found that the variability of cells improved their learning and reduced energy consumption.

Principal author Dr Dan Goodman, also from Imperial’s Department of Electrical and Electronic Engineering, said, “Evolution has given us incredible brain functions, most of which are only beginning to understand. Our research suggests that we can learn vital lessons from our own biology to make AI work better for us. ”

Modified timing

To complete the study, the researchers focused on adjusting the “time constant,” that is, how quickly each cell decides what it wants to do based on what it wants to do. make the cells connected to it. Some cells will decide very quickly, just looking at what the connected cells just did. Other cells will react more slowly, basing their decision on what other cells have been doing for some time.

AI can be compared to how our brain works by emulating certain brain properties. Nicolas perez Department of Electrical and Electronic Engineering

After varying the time constants of the cells, they tasked the network with performing benchmark machine learning tasks: classifying images of clothing and handwritten numbers; recognize human gestures; and to identify spoken numbers and commands.

The results show that by allowing the network to combine slow and fast information, it was better able to solve tasks in more complex and real contexts.

When they altered the amount of variability in the simulated networks, they found that which worked best matched the amount of variability seen in the brain, suggesting that the brain may have evolved to have just the right one. amount of variability for optimal learning.

Nicolas added: “We have shown that AI can be compared to the functioning of our brain by emulating certain brain properties. However, current AI systems fall far short of the level of energy efficiency found in biological systems.

“Next, we’ll look at how to reduce the power consumption of these networks to bring AI networks closer to performing as efficiently as the brain.”

This research was funded by the Engineering and Physical Sciences Research Council and Imperial College President’s Doctoral Fellowship.

“Neuronal heterogeneity promotes robust learning»By Nicolas Perez-Nieves, Vincent CH Leung, Pier Luigi Dragotti and Dan FM Goodman, published on October 4, 2021 in Nature Communication.

Hand gesture image: Perez-Nieves et al.

Neuron Image: Shutterstock