Monday, November 16, 2015
Didn't need experiments
Headlines:
Single Artificial Neuron Taught to Recognize Hundreds of Patterns
Biologists have long puzzled over why neurons have thousands of synapses. Now neuroscientists have shown they are crucial not just for recognizing patterns but for learning the sequence in which they appear.
Why were you puzzled? Dendrites have different lengths and go to different spots on the skin or retina or whatever. Therefore some patterns of touch or light will reach threshold and fire the axon of this particular neuron. Other patterns won't push enough dendrites to trigger. Others will be affected by decay times and refractory periods within the longer dendrites.
Tack on another basic fact, that junctions which get used often tend to strengthen while junctions that don't get much 'current' tend to fade, and you've got a trainable pattern recognizer for very small patterns. Follow this up through the layers of condensation, and you have different responses to different patterns.
All of that is pretty much obvious from "reading the schematic." The system that truly needs more research and experimentation is the active feedback from center to periphery. This has been seen but not fully understood in the cochlea, and is just now being seen in the retina. Central control sends signals to elements in the cochlea that are more like muscles than nerves, and these elements seem to physically alter the configuration of the cochlea.
¶ 4:49 AM 
