3
$\begingroup$

Most of the sources studying retina seem to suggest that retinal connectome is hardwired and species dependent (like specialized vertical movement detector in small rodents). I couldn't find a single publication exploring possibility of learning within retina. My understanding of learning is somehow rewiring ganglion-amacrine neural circuits based on which visual features are useful for a particular individual.

Modern artificial neural networks use backpropagation to learn feature detectors but there seems to be no information flowing from brain back to retina. I guess that would mean that backpropagation isn't possible in biological visual system. That in itself doesn't rule out other mechanisms of learning.

Are there grounds to believe that retinal wiring is immutable or it is just lack of research on the topic?

$\endgroup$
  • 2
    $\begingroup$ Are you talking neuroplasticity in general, or 'learning'? The latter is a higher level (cortical) process, while neuroplasticity is a very general term that even applies to a strengthening of synaptic connections anywhere in the nervous system. The former doesn't happen peripherally (a brain is needed) while the latter undoubtedly happens - one google search on 'synaptic strength alteration in the retina' will very likely turn op a 'yes' somewhere in some article or website. Could you specify what you are after? $\endgroup$ – AliceD Oct 14 '17 at 19:20
  • 2
    $\begingroup$ I meant 'extreme' neuroplasticity - strengthening/weakening of synaptic connection that affects qualitatively the kind of visual features relayed back to the brain. I guess my question makes the most sense in context of recent advancements in computer vision. It is now possible to train a multilayer neural network by feeding it raw pixels and have it automatically learn how to interpret local features like edges or blobs of color without doing any hardwiring. It would seem that neural structures within retina are doing exactly that - detecting visual features. $\endgroup$ – Andrew Butenko Oct 14 '17 at 20:26
  • $\begingroup$ Makes sense. Thanks for that. Please consider adapting our question accordingly. +1 $\endgroup$ – AliceD Oct 14 '17 at 21:25
  • 1
    $\begingroup$ @AliceD "learning" is not necessarily applied only to higher areas / processes in neuroscience, as far as I know. Change in signal processing in the retina level for "better" vision can be called "learning" in some sense. $\endgroup$ – Memming Oct 15 '17 at 19:05
4
$\begingroup$

It is not immutable. There are many studies on retinal plasticity, for example using laser coagulation and showing that after selective lesions new connections can be formed even in the adult retina, restoring the responses of retinal ganglion cells in the lesioned area (e.g. see this paper).

There is also evidence that visual experience is required for the development of the retinal circuitry after birth (see this other paper), so it doesn't seems to be entirely hardwired.

$\endgroup$
  • 1
    $\begingroup$ On a side note: afaik back-propagation is generally not considered a biologically plausible learning algorithms. Others, typically based on Hebbian learning (e.g. Boltzmann machines), are considered more plausible because they use signals that are locally available to single neurons and don't require the propagation of error values. $\endgroup$ – matteo Oct 17 '17 at 23:23
  • 2
    $\begingroup$ Also (after reading your comment below the question) keep in mind that the retina only detect local luminance contrasts using a center-surround receptive fields, it does not detect edges. Edges are processed only later in the visual system hierarchy (after LGN, in the primary visual cortex, or V1). $\endgroup$ – matteo Oct 17 '17 at 23:23
  • $\begingroup$ Huge props for the publication pointers! I didn't know that retina doesn't detect edges(I am very new to the topic). I vaguely remember theory about retinas processing being mainly about subtracting out global motion caused by eye movement. $\endgroup$ – Andrew Butenko Oct 20 '17 at 3:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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