Take the 2-minute tour ×
Biology Stack Exchange is a question and answer site for biology researchers, academics, and students. It's 100% free, no registration required.


I am a computer programmer who is fascinated by artificial intelligence and artificial neural networks, and I am becoming more curious about how biological neural networks work.

Context & what I think I understand

In digesting all I have been reading, I am beginning to understand that there are layers to neural networks. A front-line layer of neurons may receive, for example, a visual stimulus such as a bright light. That stimulus is taken in by the front-line neurons, each of which produce a weighted electro-chemical response that results in a binary decision to pass an electrical charge through its axon to the dendrites of the tens of thousands of neurons to which it is connected.

This process repeats through layers channeling the electrical signals and focusing them based on their permutations until ultimately a charge is passed to a focused response mechanism such as the nerves that control shrinking of the pupils.

Hopefully I got that correct.

Preamble to the question ;)

Assuming that I am not completely off-base with my basic understanding of how a biological neural network operates, I am beginning to grasp how an input (stimulus) results in an output (response) such as motor movement or reflexes. That would just seem to be basic electricity of open and closed circuits.

HOWEVER, what confuzzles me still is how a memory is stored. The analogy to an electrical circuit breaks down here, for in a circuit I can't really stop the flow of electrons unless I dam up said electrons in a capacitor. If I do that, once the electrons are released (accessed), they are gone forever whereas a memory endures.

So. . .

How the heck are memories constructed and stored in the human brain? Are they stored in a specific region? If so, where?

share|improve this question
Regarding pupil size alterations it is the circular muscle of the iris which controls pupil size. Effectors are always either muscles or glands - not nerves. Otherwise spot on :) –  Rory M Jan 23 '12 at 18:11
“the analogy to an electrical circuit breaks down here” – it’s not really an analogy, and it doesn’t break down. This is how the brain works: not an analogy, but rather the actual mechanism. The only part missing is the dynamics in structure. –  Konrad Rudolph Feb 26 '12 at 23:36

3 Answers 3

up vote 22 down vote accepted

Unfortunately, we are all still "confuzzled" by how memory works. We are far from a complete understanding of how memory is stored and recalled. Nonetheless, we do know a little, so read on.

Your understanding of basic neural function is almost correct. First, an individual neuron will signal through its single axon onto the dendrites of many downstream neurons, not the other way around. Second, I am not sure what you mean by "focusing them based on their permutations," but it is true that neural information can undergo many transformations as it propagates through a circuit. Third, if there is a behavioral outcome of the network activity like a muscle response or hormone release, those effects are mediated by nerves communicating with muscles and hormone-releasing cells. I'm not sure if that is what you meant by "focused response mechanism."

Finally, as you have discovered, the analogy of neural circuits to electrical circuits is relatively poor at any reasonably sophisticated level of analysis. My opinion is that biological systems are often poorly served by being framed as engineering problems. Others will disagree with that, but I think understanding a biological system on its own terms makes many things much clearer.

The key thing missing from the electrical circuit analogy turns out to be one of the keys to understanding information storage in neural circuits--the synapse, the site where one neuron communicates with another. The synapse transforms the electrical signal from the upstream neuron into a chemical signal. That chemical signal is then converted back into an electrical signal by the downstream neuron.

The strength of the synapse can be adjusted in a long-term way by changing the level of protein expression--this is called long-term potentiation (LTP) or long-term depression (LTD). LTP and LTD therefore can regulate the ease with which information can flow along a particular path. As a basic example (that should not be taken too seriously), imagine a set of neurons that represents "New York City" and another set of neurons that represents "My Friend John." If you then happen to be in New York City with your friend John, both of those groups of neurons will be active and synapses between these two networks will be strengthened because they are co-active (see Hebbian plasticity). In this way, the idea of NYC and the idea of John are now bound together.

Where are these neurons that represent NYC and John? We are still not totally clear on this, and the question is complicated because there are many different types of memory. For instance, your memory of how to ride a bike (procedural memory) is not treated the same as your memory of what you ate for breakfast (episodic memory). However, a best current answer is that the hippocampus and its associated regions are important for the initial encoding of memories and the neocortex is where longer term memories are stored. There is substantial communication between these two areas so that memories can be effectively adjusted over time.


In response to Jule's comment asking for some resources, I realize it is important to make the point that the Hebbian model I outlined hasn't been definitively shown. Like with all aspects of neuroscience, there is a lot of good work at the molecular and cellular level and good work at the behavioral level, but the causal link between the two is not so clear. Nonetheless, Hebb's idea is still the mainstream working model for how memory works. Some reading might include:

1) Neves, G., Cooke, S.F., Bliss, T.V.P., 2008. Synaptic plasticity, memory and the hippocampus: a neural network approach to causality. Nature Reviews Neuroscience 9, 65–75. A review on hippocampal memory and its relation to LTP/LTD and Hebbian theory. Notes the general difficulty of proving the theory and some ways for experiments to move forward.

2) Lisman, J., Grace, A.A., Duzel, E., 2011. A neoHebbian framework for episodic memory; role of dopamine-dependent late LTP. Trends in Neurosciences 34, 536–547. A review proposing an elaboration of the Hebbian model that includes neuromodulatory influence on plasticity and memory process.

3) Johansen, J.P., Cain, C.K., Ostroff, L.E., LeDoux, J.E., 2011. Molecular Mechanisms of Fear Learning and Memory. Cell 147, 509–524.. An excellent review on fear learning and memory with an extensive section on Hebbian theory.

4) Liu, X., Ramirez, S., Pang, P.T., Puryear, C.B., Govindarajan, A., Deisseroth, K., Tonegawa, S., 2012. Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature. A research article which is perhaps a realization of some of suggestions in the Neves et al review. They use light to reactivate a fear memory. This suggests that activation of the hippocampal network that was active during memory formation is sufficient to elicit the memory.

share|improve this answer
Thank you for an excellent answer! –  David Jan 23 '12 at 19:46
Do you have a reference to a paper that summarizes the experimental evidence for this theory? –  Jules Jun 6 '12 at 19:51
Thank you once again yamad for your thorough answer on this fascinating subject. –  Matthew Patrick Cashatt Jun 7 '12 at 18:08
For a little more on ways in which the nervous system is plastic, see my answer here biology.stackexchange.com/a/1359/72 –  yamad Nov 28 '12 at 17:35

I would like to point out some ways your understanding is wrong. "Neural networks" are usually a computer science term, only very, very loosely based on actual neural networks. The idea of layers in a neural network is pretty much an invention of computer science, it doesn't really reflect the reality. Also, neurons are not binary switches. It isn't so much about on/off, so much as a temporal rate code of action potentials. The actual brain sort of stochastic, and does not really work on the level of the single action potential.

The adaptation of the term "neural networks" by those devious computer scientists is unfortunate. Their work is virtually unrelated to actual neural networks.

share|improve this answer
Thanks Preece. That clarity is also helpful. –  Matthew Patrick Cashatt Feb 26 '12 at 14:41
I think you overstate the difference from neural networks. For instance, they are often not binary but rather have a smooth transition function (an idea copied from biology), are stochastic (an idea copied from biology) and doesn’t work on the level of a single AP. The adaptation of the term isn’t unfortunate, it’s a consequence of where the inspiration came from. Almost every CS concept has such an inspiration, and many non-CS concepts have as well (ever seen a crane?). –  Konrad Rudolph Jun 6 '12 at 21:58
@KonradRudolph They simply don't approach the complexity of an actual neuronal network. Even the computational neuroscientists, who are explicitly trying to model neurons, fall short. It's an unfortunate adaptation of terms because now "neural network" no longer refers to a network of neurons. It refers to a crude simulacrum of a neural network. The degree to which computer scientists have neglected the biology is disturbing and has stunted the field of AI. –  Preece Sep 18 '12 at 0:20
@Preece I agree with all of your comment except the last sentence. Biology isn’t being neglected, it’s simply not relevant: neural networks in CS solve a problem, and that problem isn’t “try to model biology as accurately as possible”, nor is it trying to create a “classical” AI. It’s simply a statistical tool for pattern matching / clustering. That said, modelling the biology accurately would be interesting in its own right; it isn’t done for one simple reason: it’s not computationally feasible. –  Konrad Rudolph Sep 18 '12 at 8:55
@KonradRudolph there is more to looking at the Biology then trying to model the nervous system. Neural structures are implementations of algorithms. I think analyzing these algorithms is at the heart of AI research. You don't need to model neurons to model neural algorithms. Trying to get AI without looking into them will (has) provided many interesting things, but the potential of the field is barely tapped. –  Preece Sep 18 '12 at 10:02

I can make a rough analogy in terms of digital media storage.

Our memories exist as a relationship between our perceptions and our sensations. Computers store input readily. However, humans store memories perceptually. This means who we are and how we remember an event permanently changes our recollection.

If you look at the progression of lossy video codecs as resolution has increased you'll see the algorithms have changed. Each generation has become accustomed to seeing certain types of video artifacts: from film, to analog static, to vhs, to h.264. Think of the algorithms as individual modes of perception.

We have ancient basal circuits which allow visual and frontal cortex areas to collaborate in the formation of memories. It's not an objective process. We all have our own algorithms.

share|improve this answer
Another very interesting post from you. Thanks! –  Matthew Patrick Cashatt Aug 8 at 15:55

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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