Purdue University researchers claim “organismic learning” from material that remembers how to forget
One of the key tricks the human brain uses to learn new material is to selectively forget unimportant information that it had retained previously. Electrical and computer engineers at Purdue University in Indiana now claim they have developed a system that uses a novel material to mimic this phenomenon, creating an effect they term “organismic learning”.
“The human brain is capable of continuous lifelong learning,” said research leader Professor Kaushik Roy, “and it does this partially by forgetting some information that is not critical. I learn slowly, but I keep forgetting other things along the way, so there is a graceful degradation in my accuracy of detecting things that are old. What we are trying to do is mimic that behaviour of the brain to a certain extent, to create computers that not only learn new information but that also learn what to forget.”
Working with engineers from MIT, Rutgers University, and Brookhaven and Argonne National Laboratories, Roy’s team built devices they call “organismoids” using the ceramic material samarium nickelate. “These devices possess certain characteristics of living beings and enable us to advance new learning algorithms that mimic some aspects of the human brain,” Roy said. “The results have far reaching implications for the fields of quantum materials as well as brain-inspired computing.”
Samarium nickelate displays some unusual quantum behaviour. When exposed to hydrogen, it is said to “breathe”, expanding in volume as hydrogen atoms are absorbed into its atomic lattice and contracting when they are removed. “The main thing about the material is that when this breathes in hydrogen there is a spectacular quantum mechanical effect that allows the resistance to change by orders of magnitude,” said Purdue materials engineering professor Srinam Ramanathan. “This is very unusual, and the effect is reversible because this dopant can be weakly attached to the lattice, so if you remove the hydrogen from the environment you can change the electrical resistance.”
This behaviour arises from the way that hydrogen atoms are split as they are absorbed into the ceramic lattice. Ionising into a proton and an electron, the electron attaches to the nickel atom in the ceramic, turning it from electrical conductor into an insulator. But this process is reversible, and in a paper in Nature Communications, Roy, Ramanathan and colleagues show how the extent of this conduction – insulation cycle can be tuned. The changing conductance and decay in that conductance over time is similar to a psychological phenomenon known as habituation, Roy explained.
“If I see certain information on a regular basis, I get habituated, retaining memory of it. But if I haven’t seen such information over a long time then it slowly starts decaying,” he said. “So the behavior of conductance going up and down in exponential fashion can be used to create a new computing model that will incrementally learn and at same time forget things in a proper way.”
The team terms this “adaptive synaptic plasticity” and have used it to model an effect that is causing problems in brain-mimicking (neuromorphic) computing: catastrophic forgetting, which occurs when computers try to learn several sequence of information, they will often forget previous sequences. Roy explained that this is like a human learning faces but forgetting each previous face to learn each new one.
The team believes that these systems might work with conventional electronic computing techniques, which are well suited for performing mathematical calculations but are very limited at reasoning and human-like decision-making. In particular, they might be useful in the field of spintronics, where the quantum spin direction of electrons represents digital ones and zeros. A single spintronics device might be able to mimic the properties of biological neuron and synapse, whereas several conventional electronic components would be needed to produce the same effect.
The team now intends to achieve the habituation property in an integrated circuit rather than by using hydrogen gas.