Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
×
Programming Software Science Hardware

Researchers Create Simulation Of a Simple Worm's Neural Network (tuwien.ac.at) 75

ClockEndGooner writes: Researchers at the Technische Universitat Wein have created a simulation of a simple worm's neural network, and have been able to replicate its natural behavior to completely mimic the worm's natural reflexive behavior. According to the article, using a simple neural network of 300 neurons, the simulation of "the worm can find its way, eat bacteria and react to certain external stimuli. It can, for example, react to a touch on its body. A reflexive response is triggered and the worm squirms away. This behavior is determined by the worm's nerve cells and the strength of the connections between them. When this simple reflex network is recreated on a computer, the simulated worm reacts in exactly the same way to a virtual stimulation -- not because anybody programmed it to do so, but because this kind of behavior is hard-wired in its neural network." Using the same neural network without adding any additional nerve cells, Mathias Lechner, Radu Grosu, and Ramin Hasani were able to have the nematode simulation learn to balance a pole "just by tuning the strength of the synaptic connections. This basic idea (tuning the connections between nerve cells) is also the characteristic feature of any natural learning process."
This discussion has been archived. No new comments can be posted.

Researchers Create Simulation Of a Simple Worm's Neural Network

Comments Filter:
  • by Anonymous Coward on Friday February 09, 2018 @02:05AM (#56093795)
    I, for one, welcome our simple worm AI overlords.
  • by aviators99 ( 895782 ) on Friday February 09, 2018 @02:08AM (#56093815) Homepage

    I even RTFA and can't tell. This seems a remarkable accomplishment.

    • Neither can I. The article links to the article at Google Docs [google.com] and that one links to a Youtube video [youtube.com]. Wait, it says they used RoboSchool RoboschoolInvertedPendulum-v1 environment, whatever that is. The quotation for that one is

      [11] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal policy optimization algorithms. CoRR, abs/1707.06347, 2017.

    • by Hadlock ( 143607 )

      Pretty sure this is the open worm project, they had a github project at one point

    • by Anonymous Coward

      This sounds like a project that is a few years old making a little more progress. It takes multiple servers to simulate the interactions of just over 300 neurons. This worm always has the same number of cells/neurons (forget the name for it). They simulate the connections of the neurons, but not what the neurons do because no one is exactly sure what they do.

      It sounds like they implemented something for the half they skipped before and got something reasonable for output. Others are posting that neural

      • by mikael ( 484 )

        Neurons integrate the sum of all inputs. Some increase the chance of the neuron firing. Others reduce that probability. Neurons even remember the synapse connections they made with other neurons. We can figure out what visual neurons do, by working out the pattern that generates the maximum output.

        Most critters have a CPG (central pattern generator) that generates the overall pattern for muscle movement in order to achieve some basic goal like move forwards, backwards, turn left or right. Waves of motion mo

  • #NotAllWorms (Score:5, Informative)

    by Anonymous Coward on Friday February 09, 2018 @02:23AM (#56093851)

    they took the neural net (without the weights & biases) of a worm, but modified/trained all its parameters. it's not really related to a worm at this point. I guess all this shows is that you don't need a lot of neurons to build useful systems.

    • by Anonymous Coward

      and as far as I can tell from the article, the summary is wrong - they did not create a simulation of a worm's brain. They only made the neural net "balance a stick"

      • First paragraph yo.

        C. elegans is the only living being whose neural system has been analysed completely. It can be drawn as a circuit diagram or reproduced by computer software, so that the neural activity of the worm is simulated by a computer program.

        They took the simulation of the worm's brain and trained it to balance a stick. They didn't build a neural network from the ground up with the sole intent of balancing a stick.

        • First paragraph yo.

          C. elegans is the only living being whose neural system has been analysed completely. It can be drawn as a circuit diagram or reproduced by computer software, so that the neural activity of the worm is simulated by a computer program.

          They took the simulation of the worm's brain and trained it to balance a stick. They didn't build a neural network from the ground up with the sole intent of balancing a stick.

          Yup, in fact the FA says:

          This behaviour can be perfectly explained: it is determined by the worm’s nerve cells and the strength of the connections between them. When this simple reflex-network is recreated on a computer, then the simulated worm reacts in exactly the same way to a virtual stimulation – not because anybody programmed it to do so, but because this kind of behaviour is hard-wired in its neural network.

          While narrow-AI that can totally trash any human at chess or something but sucks at everything else is interesting for limited applications it is neural networks that is the future of AI. It will be a long time before it is possible to design a set of algorithms that can outperform the result of 4,2 billion years of evolution. If you can simulate the neural network/brain of an insect or something of that level of sophistication, complete with its ocular and audio sensors you have th

          • Re:#NotAllWorms (Score:4, Insightful)

            by religionofpeas ( 4511805 ) on Friday February 09, 2018 @07:40AM (#56094441)

            I can't wait to see your trained ant drive a car.

          • by mikael ( 484 )

            We can train honey bees to play a version of bug football simply by showing them videos of watching other bees push a ball into a hole and getting a sugary reward. It was thought they just used optic flow to navigate, but that proved they could understand the location and orientation of other bees. They get by with maybe 1 million to 3 million neurons.

        • What they did not do is solve the real problem of understanding how C. Elegan's nervous system actually works. The connectome has been known for a long time. And building some sort of net with connectome is not interesting.

          Any article which does not distinguish between Artificial neural networks and real neurons is bullshit. The latter are much more complex individually.

    • Re:#NotAllWorms (Score:4, Interesting)

      by Anne Thwacks ( 531696 ) on Friday February 09, 2018 @03:49AM (#56094027)
      In nature, neural nets are dynamically rewired.

      Think of the net starting with all potential connections. the neurones sum their inputs in a weighted manner.

      Over time, some connections get stronger, and others weaker. Eventually some die out completely.

      Some of you may see this as entirely and completely analagous to how the logic and routing are programmed in FPGAs.

      At night, the whole shebang goes off-line for a bug fix session (Some of you may see this as entirely and completely analagous to how FPGAs are put in programming mode for the logic and routing to be programmed).

      Some of you may think that either:

      • A lot of these "neural net" research projects are a scam, or
      • A lot of these "neural net" research projects are a serious waste of time, or
      • The people involved in these projects might be better off reading up on FPGAs before they set out to prove the blindingly obvious, rather badly.

      Yes, I have been saying this since Xilinx released their first chip.

      Yes, it would appear that an FPGA is more intelligent than 90% of politicians. Appearances can be deceptive. Its probably more like 99%

    • As this is not a generalized neural net simulation but a specific one whose connections both between neurons and from neurons to sensors and actuators mirror the real world one, it is likely that they had to achieve weights and biases very close to those in the real world to get it to work. The exact same neurons would have to be actuating things in response to the same neurons sensing things as in the real world.

      I see this as a validation that there are few remaining surprises in how these neurons work.

      Tha

  • by Anonymous Coward

    At first glace, while distracted, the title looked like it said "Researchers Create Simulation Of a Simple Woman's Neural Network".

  • by Anonymous Coward on Friday February 09, 2018 @02:48AM (#56093909)

    ... it was the lovely folks at the Technische Universität Wien. Though I am quite sure they would welcome a glass of Wein (wine) every once in a while too.

  • by Guppy ( 12314 ) on Friday February 09, 2018 @04:29AM (#56094105)

    It just so happens that C. elegans is one of the few multicellular organisms for which all cell fates during development are mostly deterministic and completely known.

    The actual worm itself has exactly 302 neurons, and their connections have been mapped.

    http://www.wormatlas.org/herma... [wormatlas.org]

  • Wein?!? like wine? Are you drunk?

  • Can we say for definite that this worm isn't a conscious being based on this research?
  • FTA:
    "But no human being has written even one line of code for this controller"

    Umm, excuse me? Its a simulation running on a computer, of course someone wrote some code - probably quite a lot of it in fact for both the training and the actual running even if these researchers themselves just used some off the shelf library such as tensorflow.

    ANNs are simply code running processing data whatever the high level logical view of them may be. The fact that the data (weights and thresholds) itself is the main driv

    • "Anyway, neurons are simply complex highly interlinked analogue logic gates - given enough time it would be possible to codify any neural network in boolean logic using thousands or millions of if-then statements and jumps."

      While true, the accomplishment is in actually doing it. Saying such as such as theoretically possible is nice and all, but actually doing it is where it is at.

      Now that they've done the worm and seen it is successful, hopefully the work continues to more and more complex organisms. They'l

  • by Anonymous Coward

    C. Elegans has been neurally simulation mapped for quite some time, and some engineers have already tied the neurons to actual physical sensors and motors to create a robot. I would really like to know how this simulation differs from the prior full neuron simulations.

  • by 110010001000 ( 697113 ) on Friday February 09, 2018 @07:37AM (#56094435) Homepage Journal
    Computer Neural networks work nothing like the human brain. Even calling these programs "neural networks" is misleading. Saying you have created a "simple worms neural network" is a complete lie. It implies that they have replicated worm level brains in the computer. They havent'.
  • But is it small enough to use for fishing?

  • by Anonymous Coward

    That's very cool!

  • They did this with roundworms already, and put it on a Lego [sciencealert.com] robot.
  • Lobsters are the next step.

Keep up the good work! But please don't ask me to help.

Working...