Looking at neural networks, each node can be described as a neuron with either several long input wires, or several long output wires.

A neuron, however, has many short input wires (dendrites) and a single long output wires (axon). Thereby not fitting either case.

How do we reconcile these difficulties?

  • $\begingroup$ Typical neurons have many inputs and many outputs. Same with neural networks. There is no such thing as length of connections in neural networks. There are lots of reasons neural networks are not perfect models for real brains, but I don't understand your description of a difficulty. $\endgroup$
    – Bryan Krause
    Sep 15, 2019 at 1:51
  • $\begingroup$ @Bryan well just looking a a neural network how it is set out, yes, technically you can arrange it in any shape, but in any way you arrange it, the typical model of the neuron doesn't seem to fit. $\endgroup$
    – zooby
    Sep 15, 2019 at 17:25
  • $\begingroup$ I think you misunderstand the purpose of neural networks and the actual structure of real neurons, though. You are misunderstanding the "long axon" model entirely. $\endgroup$
    – Bryan Krause
    Sep 15, 2019 at 17:47

1 Answer 1


A: Not all neurons have many short dendrites and a single long axon. In fact, the majority of them do not have this shape. Cerebellar granule cells, which are the most numerous neurons in the brain (estimates of their total number average around 50 billion, which means that they constitute about 3/4 of the brain's neurons)[Ref 1,2,3], do not have this shape. They have an axon that branches out into long parallel fibers as shown in the diagram below (modified from the figure in Ref 2). enter image description here The shapes of neurons vary greatly. This is because neurons have evolved their structures to suit their functions:

We can learn a lot about what a neuron does by looking at it’s morphology (i.e. shape). For example a neuron with large branching dendrites is likely integrating information from a large number of inputs, whereas a neuron that has dendrites that branch close to the soma, but don’t extend very far, is probably only integrating information from it near neighbors. Same thing with axons, projection neurons have long axons that allow them to communicate with neurons in distant brain regions, while a local interneuron will a short axon that will only branch locally, allowing it to talk to nearby cells. Some cells in the peripheral nervous system have the axons coming directly out of the dendrites, allowing them to efficiently convey information from one to the other. [Ref 4] (The figure below is also from Ref 4)

enter image description here

So, neurons have evolved their various shapes to do their functions. I think whether the one shape of these, as questioned in this thread, or many or all of these shapes are right or wrong for us to use to model neural networks is not answerable now. But I think we should try to study and learn from them because it is the best model we have. Remember, these neurons with their various shapes and connections are successful in creating one of the most wonderful phenomena in this universe: our conscious mind.


  1. Richard H. Masland. Neuronal cell type. Current Biology. 2004; Vol 14 No 3: R497-R500.

  2. Erik De Schutter. Glomeruli of Cerebellar Cortex: Computation by Extrasynaptic Inhibition?

  3. Wikipedia. Cerebellar Granule Cell.

  4. Masland RH. Looking at Neurons Brown University. Neuroscience in Action: Understanding Our Brains and Nervous System.

  • $\begingroup$ Interesting. I wonder why we are generally only taught one model of the neuron with short dendrites and a long axon. $\endgroup$
    – zooby
    Sep 15, 2019 at 17:23
  • $\begingroup$ I think that’s because this model can teach all the important parts of a neuron (dendrite, soma, and axon) fairly completely and easily. Also, although it is not the most common shape of neurons in our brains, many important types of neurons have this shape, such as motor neurons in the brain and spinal cord, ascending sensory neurons in the spinal cord, and descending autonomic neurons in the spinal cord and peripheral nerves. So, they are probably the most familiar shape of neurons that we have dealt with. $\endgroup$
    – user287279
    Sep 16, 2019 at 1:18
  • $\begingroup$ Oh, btw, perhaps, what we can learn from these various shapes of neurons is that, maybe, to achieve the better or best results, we have to model different types of circuits in neural networks to best suit each functioning module? $\endgroup$
    – user287279
    Sep 16, 2019 at 1:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.