Neuronal Learning Observed

Gregg Favalora writes "According to this week's EE Times, R. Colin Johnson reports that researchers at UC San Diego have directly observed the physical changes that neurons undergo during learning. His article explains that neurons were cultured on a smooth, photoconductive silion substrate. Using optical techniques, they were able to trigger individual neurons into firing -- and were actually able to observe some of the physical changes that underly short- and long-term learning. According to the article, "[The team] tested out the theory that learning results from a physical change that strengthens the connections between selected neurons. [They showed] how short- and long-term memories result from different physical effects in the brain. Short-term memories, it turns out, result from the instant assembly of more filaments to strengthen the skin of the cell temporarily, whereas long-term memories result from the growing of a new synapse to strengthen the connection permanently." Besides the interesting cellular observations they're making, I am also intrigued by the process the article describes which uses properties of the silicon substrate to aid in firing individual neurons. "

Comments

This is really an important discovery

[3ryon]

For more background info on how neurons work: see How stuff works sub-page on the brain. I was hoping that they would have a good discussion on how sodium and potasium ions move through the cell membranes creating a charge, but at least it's a good intro.

It's the other way around!

[itwerx]

You're right on the Johnny Mnemonic thing, but hey, if I could plug a module in my head that stored an entire encyclopedia and be able to access it like it were an adjunct to my own memory think what a benefit that would be in so many environments!!
Need to repair that aircraft engine you've never seen before? Plug the chip in! S'long as one has a foundation of basic skills and knowledge, the minutiae of many fields could be placed on a chip, saving (potentially) years of study.
If we could interface a neuronal structure (our brain) to a silicon structure (ROM) it would totally revolutionise the way we conduct our entire existence.
Not to mention the potential for interfacing. (DOOM the way it was meant to be! And other, er, "entertainment" :)

Institute for Neural Computation Homepage.

[vtechpilot]

For the lazy:

See: http://www.inc.salk.edu/

The mechanical guides the logical

[adamy]

That the mechanical guides the logical has been the basic assumption of Neural networks research for a long time. By Mechanical, I mean the physical connection between neurons, as opposed to the chemical levels in the neuron (the neuron holding state).

I've read that neurons can feed back into themselves, kind of like latches in computer memory, but in a much more complex way. I wonder if this is how the brain knows how to do long sequences: Part of the neural net keeps the brain focused on the task at hand, Say playing a song on a piano. The combination of the steady state and the current state ( I am at measure 4, third note, held for a count of 2) Figures out what to play next (G major chord in the left hand, start the trill with the right.)

So to learn a long sequence, the brain must start off with the short term memory of reinforcing with fibers the synapses for certain combinations...and then make new connections. That is why it is hard to learn a new song, and possible to play something you memorized in 5th grade. But since the actual playing of the instrument is common to both, it fades into the background.

One concept that I read about that is helpful in the study of Neural Nets is Orthogonaity. The more different two things are from each other, the easeir they are to differentiate. IE, Fire either Neuron 1 or Neuron 2 type distinctions. I guess that is why two things that are very similar (two different editors with different shortcut keystroke settings) can really confuse you...at least until the short term memory fibers kick in and reinforce the current task. Over time, It should get easier to switch between the two editors...just need to kick your brain into the right editor mode. Since Typing is the same for both of them, it fades into the back ground.

I wonder what triggers the start of the long term memory building process. Is it a threshhold of the short term memory that, once reached, kicks it into gear? Or is it a gradual process: adding more fibers will eventually build another connection.

hebb learning

[gnujoshua]

The idea that presynaptic firing and postsynaptic firing control the synaptic efficacy of a cell was proposed by Donald Hebb in 1947. However, it has also been questioned that different neurons behave differently. For instance, granular cells may act differently then cells in the lateral geniculate nucleus (they are drastically different in size and number for one). Therefore, these scientists have shown a particular case, they have not shown that Hebb learning can be generalized throughout all the brain, only that in certain cells obey Hebbian learning.