Researchers have discovered that the common blow-fly has a brain that fires neurons in a way that could be replicated by highly efficient next-generation computers.
Head of a blow-fly.
Past research - dating back to 1926 - had found that the timing of the fly's neuronal firings is quite random. But the new study shows that these experiments used stimuli that were too boring for the fly, and the most efficient way to process the boring information was through firing random signals.
In the new experiment, a collaboration of researchers from four institutions (Los Alamos, the Hun School of Princeton, Princeton University and Indiana University) performed a more natural experiment. The group harnessed a fly on a turntable-like system, which could spin very fast and quickly change velocities - very similar to the type of acrobatic flight that a fly undergoes in real life.
When spun on the turntable, the fly saw the world similar to how it would see it if it were really flying. This realism enabled the researchers to observe the firing patterns of the wired fly's motion-sensitive neurons more accurately.
They observed that, under complex flights, the fly's neurons fire very quickly, and with precise timing. The researchers were able to map the fly's firing patterns with a binary code of ones and zeroes, similar to computer code. The group described the firing patterns as a regular "language," with the neurons firing at precise times depending on the visual stimulus.
Because of the precision and regularity of its neuron spikes, the way the fly reacts to motion stimuli could enable researchers to use the fly as a model for designing future computers. Previous "neuromimetic" approaches to artificial intelligence systems have often been based on a number of impulses firing during any time within a time period. The new study shows that it may be necessary to control the precise timing of these impulses, since timing is important for the fly as it controls how it moves through the world.
"This may be one of the main reasons why artificial neural networks do not perform anywhere comparable to a mammalian visual brain," said Ilya Nemenman, physicist at Los Alamos.
Possibly, computers that take inspiration from the fly's neuronal firing patterns could improve the analyses of satellite images, facial-pattern recognition systems, and other vision-based systems.
via: Los Alamos