Scientists train a computer to learn like a human
In a now-famous experiment, Russian physiologist Ivan Pavlov taught his dog to associate food with the sound of a bell ringing, so much so that the dog would start to drool when it heard the sound. This was a demonstration of ‘associative learning’: how our experiences and memories can lead to unrelated ideas being linked in our brains over time.
Now, a team of researchers at Northwestern University and the University of Hong iong have developed a device that can learn over time in the same way, mimicking a human brain. This technology could be more energy efficient than traditional computers, the scientists say, and can even continue to run smoothly when some components fail.
The brain has synapses that connect neurons together and pass messages between them using chemicals called neurotransmitters. Strengthening the connection between two neurons results in the creation of memory.
Inspired by this process, the team developed the ‘synaptic transistor’, an organic, electrochemical device that can trap ions. The ions behave like neurotransmitters, and trapping them allows the transistor to ‘remember’ previous activity. This process lets the device form connections over time.
The team conditioned the device to associate light with pressure. First, they pulsed an LED light bulb, which the device picked up with its light sensors, and they immediately followed that by applying pressure to its pressure sensors. After several training cycles the circuit had formed a significant association between light and pressure. It started to produce a signal corresponding with pressure after
only a ƃash of light.
“Although the modern computer is outstanding, the human brain can easily outperform it in some complex and unstructured tasks, such as pattern recognition, motor control and multisensory integration,” said Dr Jonathan Rivnay, assistant professor of biomedical engineering at Northwestern University and a senior author of the study.
“This is thanks to the plasticity of the synapse, which is the basic building block of the brain’s computational power. These synapses enable the brain to work in a highly parallel, fault-tolerant and
energy efficient manner.q