Comment neurons -- cellular automata (Score 4) 186
I've seen some basic laboratory work in a physics conference and read some theoretical works prior to this report. If you think of neurons as basic units (as they should be), what is the optimal behaviour they all should have in the beginning (birth)? This is one of the central issues of neural computing. It's now believed by many that the spike trains that neurons emit to their neighbours contains the "information content". The first thing one could do with the spike trains is to retransmit them, or return them to the senders. It turns out that it is exactly what neurons do when they first find each other out. Only things get really messy and intractable when they seem to know what they are doing. (one obvious behaviour is specialization, which could be a result of instability, or phase separation, of the syncronization process). The efforts these guys are trying are probably to exploit some known behaviour after neurons somehow begin to stabilize into some functional units.
One reason why the problem is so difficult is that information is not encoded in a static physical format. In a digital computer, you may stop the quartz oscillator and hold some gates to on or off to read out the specs, painstakingly. On a neuron, you can't do that! Spike trains are dynamical processes that have many more possible ways to encode information. A useful analogy is from languages. Let's say every single individual in this world speaks a different language in the beginning, but with the same alphabets. When I write "one" on the floor, how would the guy next to me know what it means when the word of the same meaning for him/her is "aye caramba"!
This field is a very broad subject encompassing biology, physics and statistical mechanics. One may found an interesting but quite speculative starting point to work its way backward from Frank C. Hoppensteadt et al. in the April 5 issue of Physical Review Letters, 1999. Science and Nature also may often have articles on the latest development in this field.
One reason why the problem is so difficult is that information is not encoded in a static physical format. In a digital computer, you may stop the quartz oscillator and hold some gates to on or off to read out the specs, painstakingly. On a neuron, you can't do that! Spike trains are dynamical processes that have many more possible ways to encode information. A useful analogy is from languages. Let's say every single individual in this world speaks a different language in the beginning, but with the same alphabets. When I write "one" on the floor, how would the guy next to me know what it means when the word of the same meaning for him/her is "aye caramba"!
This field is a very broad subject encompassing biology, physics and statistical mechanics. One may found an interesting but quite speculative starting point to work its way backward from Frank C. Hoppensteadt et al. in the April 5 issue of Physical Review Letters, 1999. Science and Nature also may often have articles on the latest development in this field.
