I'm new to neuroscience and psychology and recently learned about Hubel & Wiesel cat experiment where "some neurons fired rapidly when presented with lines at one angle, while others responded best to another angle" link.

That made me wonder, how are "ideas" or "concepts" appear in the brain? If a single neuron can represent a certain line angle, do several of these represent a concept of a shape? Or a house? That is, is a collection of neurons that fire at the same time a "concept" or "idea"?

I was trying to find resources on this question but didn't find many other theories.


First of all, let's look at human perception. Humans have several sensory modalities: sight, hearing, touch, etc... Note that there are many more than I just enumerated, and certainly not merely the traditionally considered 5 senses. To illustrate my point, we will look at the best studied modality: sight. Usually, external stimuli will reach specialized neurons which act as transducers: rods and cones in the retina turn light into an electrical potential. This information, which has just entered the nervous system, will sometime undergo some limited local processing near the sensory organ: once again, this is the case for the retina, as the initial transducers are connected to other kinds of neurons in the retina itself (e.g. retinal ganglion cells, RGCs). The information then eventually makes its way to the thalamus (sort of a "relay station"), from which it reaches a primary dedicated processing area. For sight, this primary dedicated processing area is the primary visual cortex V1 in the occipital lobe.

Here is where it starts to get interesting. Certain cells, named simple cells, are tuned to a specific orientation of a line in a certain area of the retina, termed the receptive field: this is what you are referring to in your question. That responsiveness takes the form of the neuron firing more rapidly the closer the orientation of the line is to its preferred orientation. These neurons detect the simplest of features of an image, just small line fragments. However, they are connected to other neurons, which respond to slightly more complex features (e.g. squares, circles, etc.), which are in turn connected to other neurons, and so on and so forth. This paradigm of emergent complexity is very similar and perhaps even more striking in the convolutional neural networks (CNNs) in machine learning, which structure their neurons on layers, each layer representing more abstract features (e.g. lines to geometric primitives to body parts to humans). CNNs were actually initially inspired by the brain, and the same phenomenon of emergent complexity can be seen in the ventral pathway of the visual system, responsible for identifying an object you are looking at. The visual signal actually goes through multiple areas after V1 (V2, V3, etc., not in strict sequential order), which each deal with different more abstract information.

You should now have a pretty good idea of how individual neurons can encode information. Unfortunately, the brain isn't that simple, and the first degree of additional complexity we can consider is groups of neurons that represent information. To illustrate this, let's look at the intention of moving a limb and thus at the primary motor cortex M1. Just like for the visual system, each neuron is tuned to a specific direction of movement, and we can once again see that they fire more rapidly as that direction of movement is matched to their preferred one. But many neurons in M1 will fire at the same time when executing a movement: how to then tell which way the limb will move? The answer is simple, but requires us to account for a group of neurons which we call a neuron population. The movement is then represented as a weighted average of the neuron preferred directions, where the weights come from the firing rate of the neurons at the time: this average is called the population vector, as shown in this seminal paper.

All good, but we're still far from done. Although firing rate coding is quite common, the nature of the neural code is still a subject of debate, and there seem in fact to be multiple neural codes. For instance, temporal coding, where the precise timing of spikes carries information, has been shown to be present in certain brain areas.

Reaching now the final step (for this answer, because we can go infinitely further), information representation can involve cell assemblies in complex ways, for instance to store and retrieve memories. A memory can be stored by a group of interconnected neurons in the pattern and strength of the synapses between them. To recall that memory, once the neurons start firing, this cell assembly will converge to a stable pattern of activation, achieving memory recall.

Resources and further learning

If the basic level of information representation in the brain which I first described is what interests you, I would suggest you look into a field called psychophysics and into perceptual learning.

If the medium complexity processes such as memory storage and recall, or decision making, interest you, I would suggest an excellent course (or its associated textbook) called Neuronal Dynamics.

Let me know if you'd like further resources in one of these directions.


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.