Pretty cool, but before anyone starts talking about the coming age of bio computation, keep in mind neurons are a horribly inefficient way of carrying out this kind of computation.
This might lead to a deeper understanding of how neural nets work (or an easier way of implementing them in the hardware, as it were). If we can design regular logic in neurons, we can (eventually?) do some of the fancy neural nets, and then we have the basis of an actually useful breakthrough.
If we can design regular logic in neurons, we can (eventually?) do some of the fancy neural nets
Well, yes, if you can make logic gates out of neurons you could eventually implement a neural network with actual neurons, but it would be little more than a perverse pun. Unless I'm missing your point...?
I was thinking something similar, that you are taking a complex biological computation device that has over 1000 synapses, each carrying some complex quantity of information, and reducing it to a bit. This is an inefficient use of a neuron.
You are quite correct; however, I think one interesting corollary to this research is the interface between brain and computer. Though we still require a method by which to translate brain function into Boolean-esque logic, at least when we figure that out we'll have a method to input that data into a computer (however inefficient that method may be).
This requires one of those "yes and no" answers. Sometimes cells auto-destruct and are replaced by new cells. Other times, they heal. Both strategies can be useful, depending on the situation.
So would the barrier to making things like adders, latches, etc. (in other words, everything you need to build a Von Neumann computer) be that they don't have a way to make NOT gates? Or would making NOT gates out what they have not be a big deal?
14 comments
[ 276 ms ] story [ 90.6 ms ] threadI'm even willing to start thinking in stack notation!
http://www.weizmann.ac.il/complex/EMoses/projects/NeuralDevi...
This shows an oscillator built of diodes. On to the Nature Physics article, now.
Cool stuff though.
Well, yes, if you can make logic gates out of neurons you could eventually implement a neural network with actual neurons, but it would be little more than a perverse pun. Unless I'm missing your point...?
The recent discovery that it may be possible for a single neuron to store a memory makes it all the more likely ( http://www.livescience.com/health/090125-memory-cell.html )