OpenWorm – Create a virtual C. elegans nematode | Hacker News Reader

OpenWorm – Create a virtual C. elegans nematode

openworm.org

51 pointsemgeee12 comments

I backed an OpenWorm Kickstarter campaign to earn a bragging right, so let me brag. :)

Seo Sanghyeon at http://www.openworm.org/supporters.html

I am one of the scientists who contributed to OpenWorm with C.elegans behavioural data. We have launched an online repository of actual experiments for anyone to download and continue our research: http://wormbehavior.mrc-lmb.cam.ac.uk/

Original paper: http://www.nature.com/nmeth/journal/v10/n9/full/nmeth.2560.h...

Also check out this embodiment

http://www.i-programmer.info/news/105-artificial-intelligenc...

And some implementation details

https://news.ycombinator.com/item?id=8745639

I find it interesting and also sobering that the connectome of the C. elegans is known since 1986 but it's still far from obvious what to do with it.

That's starting to be figured out finally. They've identified the motor neurons and some of the ones involved with touch so that a simulated worm can actually navigate around. I find it incredible though all this is done with as few neurons as it has (302). I've got everything downloaded and have been wanting to play with it in a simulation of some kind.

C. elegans neurons are non-spiking, making them more like analog circuits than digital circuits. In comparison, all mammalian brains use spiking neurons.

Interesting, I always wondered if there were differences like that. The way I'd think of that if I were translating it to circuitry would be to say that the mammalian brains operate at saturation of a transistor (fully on, or fully off). And that C. elegans would then be operating in the linear region of the transistor; assuming that you could represent each neuron as a transistor of course. I wonder if the change in structure happened roughly for the same reason that we're changing much of our circuitry from analog to digital (or at least avoiding analog as much as possible) because we can get better noise response and simpler circuit design. Makes me wonder then if something like epilepsy could be caused by some neurons operating the wrong way and not giving clear signals down the chain which could lead to feedback and/or noise that ends up detrimental.

Does this mean that one neuron can carry a signal that is more dense in data than a 0/1 binary spiking signal? Like a single neuron can transmit a complex feeling instead of just if it's touching something or not?

It can transmit continuous variable state rather than discrete state in pulses.

Of course you can still transmit continuous variable state indirectly using pulses (for instance, using pulse width modulation or frequency modulation) but it requires a bit more hardware.

Probably there is some kind of cross-over in complexity of the neural structure (density, length of the individual neurons) where it makes more sense to switch to a digital mode (cross talk for instance).

Indeed,

The question I'd have is whether the connection-set or "connectome" is sufficient to actually describe the behavior of the organism. As far as I know, a CS "neural network" is by no means equivalent in behavior to the neurons in a brain, even if each have the same connection topology/graph.

As far as I can tell, some group of neurologists have decided the best way to go is to toss topologies and rough simulations to CS and open source people 'cause neurologists don't do really complex systems. But it seems like that still doesn't solve the "this model is wildly unrealistic" problem

Blue Brain Project is working on Channelpedia, a collection of biologically accurate models of neurons. They check all models against real neurons.

http://channelpedia.epfl.ch/

http://www.openworm.org/science.html

s/pealing/peeling

Reply on news.ycombinator.com