Now, researchers at Arizona State University have come up with a model that could help unlock some of the secrets of how humans process patterns and possibly lead to smarter robots. The advance concerns oscillatory associative memory networks, basically the ability to see a pattern, store it and then retrieve that pattern when needed. A good example is how humans can recognize faces.

"It is still a really big mystery as to how human beings can remember so many faces, but that it is extremely difficult for a computer to do," said Ying-Cheng Lai, an ASU professor of mathematics and a professor of electrical engineering in the Ira A. Fulton School of Engineering.

Lai, along with former post-doctoral fellow Takashi Nishikawa (now at Southern Methodist University), and former ASU professor Frank Hoppenstaedt (now at New York University), published their research, "Capacity of Oscillatory Associative Memory Networks with Error-Free Retrieval," in a recent issue of American Physical Society's Physical Review Letters.

Although what the team developed is a mathematical and computational model for oscillatory networks that can be used associated memory devices, implementation of the model is possible by using electronic circuits as phase-locked loops.

"Computers can do very fast computation that humans cannot do, but humans can recognize patterns so much better than computers," Lai said. "The question is why. What is the fundamental mechanism that a biological system like us can make use of and try to memorize patterns."

A key to pattern recognition is the use of oscillatory associative memory networks. Lai said the human brain and its use of neurons have a great advantage over computer memory in that they employ oscillatory memory systems, systems where the individual components can oscillate or freely change between states. In contrast, digital computer memories operate on a binary number system (1 or 0).

An important advance was made in this area in the 1980s by John Hopfield, a Caltech researcher at the time, who developed the "Hopfield network" to help understand how biological memory works. But the main drawback of the Hopfield network is that while it represents how biological memory works, it employs discrete state memory units while most biological units are oscillatory.

"Our work is the first demonstration of the possibility for oscillatory networks to have the same memory capacity as for the discrete-state Hopfield network," Lai said. "When the Hopfield network was invented, it was considered a revolutionary step in understanding how biological memory works.

"A difficulty with the Hopfield network is that it consists of units (or artificial neurons) with two discrete states," he added. "It is therefore desirable to study oscillatory networks but this has been a struggle, as all previous work shows that the capacities of these networks are very low compared with that of the Hopfield network. In a sense, our work helps solve this difficulty."

Lai said that the most immediate application for this research is in artificial intelligence, where researchers try to get computers to reason as a human would. He adds that this advance could possibly allow the development of artificial memory devices that would use oscillators, which are robust and secure.

This could mean robots, or other electro-mechanical devices controlled by an electronic "brain" that could recognize patterns and do some form of reasoning on the fly -- basically respond to a much wider range of unanticipated situations -- to perform its task. This would be a big step towards smarter robots.

But the real payoff in Lai's research could be what it may provide in terms of basic research into the human brain itself. Developing a good model of the human brain, one that could more closely replicate the actual function of the brain as it reasons, might help understand more of its operational basis and how it developed into the organ it is today.

"Biological systems, such as cells and neurons, are oscillators," Lai explained. "Demonstrating that oscillatory networks can have memories with high capacity is one more step toward understanding biological memory.

"Although the classical Hopfield network provides a plausible mechanism for memory, it has the drawback that it is too idealized as compared with real, oscillatory biological networks," he added. "We hope our work will stimulate further studies of the origin of memory based systems on a more realistic oscillatory network."

Note: If no author is given, the source is cited instead.

• Memory
• Intelligence
• Psychology

• Artificial Intelligence
• Computer Science
• Computer Programming

• Computational neuroscience
• Amnesia
• Computing
• Social cognition
• Security engineering
• Computer