Oct 31 2008
Single Neuron Neuroscience
During some of my skeptical lectures I flash a picture of a familiar movie star on the screen for a fraction of a second to demonstrate that most people can recognize the person, out of the millions they have ever seen, even after such a brief exposure. It’s a quick way to demonstrate the power of our face-recognition ability, which still outstrips that of supercomputers. I use it to make the point that our brains are exceptional at pattern recognition, which also helps us understand one of our major cognitive weaknesses – hyperactive pattern recognition, or seeing patterns where they do not really exist.
Neuroscientists are also interested in face-recognition, and other pattern-recognition, that occurs in the human brain. R. Quian Quiroga from the University of Leicester has published a number of studies exploring this ability. His latest research reveals what the press has dubbed “The Jennifer Aniston neuron.” Quiroga recorded the activity in the visual cortex of using micro-electrodes of subjects viewing the faces of familiar stars, like Aniston, Tom Cruise or Halle Berry. He found that each familiar face seems to be encoded in a single neuron.
This finding is a bit surprising as it was previously thought that visual memories such as this would be stored in a network of neurons. Rather, our brains assign a single neuron to remember a single familiar face, and that neuron will fire whenever that face is viewed – in any condition and from any angle. Previous research from Quiroga looking at medial temporal lobe responses to familiar faces and object found that the response of the memory neuron is an all-or-none response, and that it can be fired even by stimulation as brief as 33 miliseconds (thousandths of a second).
This all-or-nothing finding is not surprising at all – the nervous system generally functions that way. You cannot partially fire a neuron (although the rate at which a neuron fires can alter). It also matches our common subjective experience, when trying to recognize someone the answer may seem to just click into place – recognition comes suddenly with all the attendant information in tow.
How does a single neuron store so much information? Well, those single neurons do not exist in isolation – on average they make over a thousand synapses (connections) with other neurons. Those synapses can be excitatory or inhibitory, and they can vary in strength. So while a single neuron may be the location where the memory of a specific face, for example, is stored, the abstract information about that face is likely stored in the pattern of synapses made by that neuron. Individual neurons may be part of many such groupings, reflecting the brain’s massively parrallel structure. This also explains the brain’s pattern recognition ability, as overlapping networks of neurons will allow one image to remind us of another image because they share neurons.
Understanding the role of these single master neurons, acting like the “address” of a memory, is an important incremental step in our progressive understanding of how the brain works. This research also shows that our technology is starting to get down to the resolution, both in time and space, at which the brain is functioning. At least we are getting down to the single neuron level. There is still a finer level of detail in all the synapses being made by those neurons.
![[del.icio.us]](http://images.del.icio.us/static/img/delicious.small.gif){kind=small}
Print This Post
19 Responses to “Single Neuron Neuroscience”
This work is very interesting, and has gotten a lot of press that suggests the ‘grandmother cell’ theory of visual recognition is right. This would be a bit of an overstatement. Quaraga has a paper discussing why this interpretation would be a severe overstatement here. They estimate that about two million neurons in IT are activated by an individual.
Population coding (the standard view of sensory coding) is consistent with fast reaction times. Indeed, you likely aren’t going to evoke a behavior quickly by causing a single neuron in IT to fire! Rather, there is likely a population of neurons responding quickly in IT, and this will more efficiently drive the downstream neurons. Also, it would be horrible to have one neuron represent a person: if that neuron dies, you are screwed!
However, clearly their work shows that not all neurons in IT respond to every stimulus, so we have something between a full population code and grandmother cells.
A familiar face is assigned to a single neuron in our brains? Wow! That result is very surprising to me, although I guess I’m not really qualified to be surprised by neuroscience, if you know what I mean.
I wonder, does the same single neuron fire in the same way if we are presented with a caricature drawing of a familiar face? It’s obviously not the familiar person, but it also obviously represents the familiar person, to a human brain. I wonder if caricatures work so well, because they are able to tap into our facial recognition networks by emphasizing only the geometrical features that dominate the network’s connections. I.e., maybe those guys on the street corners and at the state fairs have really been doing neuroscientific research all these years!
By the way, the word “parallel” is misspelled in your penultimate paragraph.
Jim: it’s more likely that a couple of million neurons respond to an individual’s face. See my comment above and the paper to which I linked, written by the researcher mentioned in the post. The ‘grandmother cell’ view is not viable.
Clearly there can’t just be a single neuron in the whole brain which codes for a given face. that’s implausible for a host of reasons. Not least because if there really were a single Jennifer Aniston neurone you’d need to record from every single cell in the brain in order to find it. (And you’d probably kill it in the process of recording from it…)
Surely what this research must actually show is that face representations are fairly sparse, so that each face activates a small number of neurons, and these neurons only respond to that face? But of course “a small number of neurons” is still thousands or millions
Steve, how would the new research differentiate the “single neuron whose pattern of synapses encode the information” idea from an alternative like the single neuron just being the output neuron of a network? In an “all roads lead to Rome” fashion, all Jennifer Aniston-y characteristics are processed by this network, and those that match eventually reach their way to the output neuron. If a tipping point is reached, then the output neuron fires, which is the event in the brain that recognizes the pattern as Jen’s face. I think what I said is closer to what Eric just said, but still not the same.
And whatever’s behind the black box, would it imply that if a single neuron is ultimately responsible, and that neuron dies, then we lose our ability to recognize that person’s face? I think you talked about something similar to this, but it was WRT the emotional impact that accompanies the facial recognition, and not the recognition itself (so they might believe their loved one has been replaced by an imposter).
Some people reading this might wonder how you get single neuron recordings. The answer is simple, really thin wire.
So is there a difference then, when I look at someone and know they are familiar, but cannot immediately put a name to the face?
This happens often, and I will have a few flashes in my memory of the person, and the more I think on it, the more I can glean from the memory; placing the person at some point of time in my life.
The name may come, or it may not. It is not always instantaneous though.
wastrel – I don’t know. The questions is – is the vague sense that a face is familiar activating the neuron discovered by Quiroga, or does it just seem familiar because of vague pattern recognition in the neurons encoding abstract details of the face, and then if you remember who they are that represents the one neuron firing?
I need more details of Quiroga’s latest study (has not been published yet), and I suspect we need to do more research to answer that question.
It seems, though, in the current study that they just measured activity when the face was flashed. Also, prior research of Quiroga suggests that the memory neurons are firing even before conscious awareness of recognition. I need to dig deeper into this research, and my current sense is that the research needs to progress some as well.
Thanks, Eric. That report was helpful.
“The Jennifer Aniston neuron.” Sounds like a way to sell papers.
“The god spot,” was good for a while too.
It seems the research here to comes after the paper Eric has a link to. (Excelllent paper- thanks Eric)
This research (as far as I can tell) deals with familiar faces and objects.
I often will see someone and not get a name for sometime afterwards (maybe even a day) Sometimes I get a vague- ‘boy that person looks familiar’.
But there are those people that I am very familiar with or who I have deemed very important that I recognize right away.
It would seem that there could be two different types of recognition pathways for faces (as well as other objects).
Very interesting stuff. The human brain always continues to fascinate me.
I’ve often wondered if “subconsciously” humans developed computers to work in the way they do because it mirrors the way our brains work…
I realize this is making a lot of grand assumptions and the whole correlation vs. causation thing, but it’s an interesting thought… in my opinion.
@ godkillzyou: our computers do not really work at all like our brains do. Computers use basic, hard-wired binary logic, each neuron in our brain is a complex non-linear self-modifying analog calculation device. On the opposite extreme, each area of a healthy brain has a specific, fairly fixed task while computers are very generic and able to run any sort of program that is loaded into them. I really have some trouble seeing any similarity between our brains and computers, either in what they do or how they do it.
I agree with TBC, the computers we use and familiar with are completely different than the neural networks we think with (which we are not even close to understanding in detail).
Rather there being a single cell responsible for recognizing Jennifer Aniston or Halle Berry (as Quiroga et al point out, this is implausible for a number of reasons), it is more likely, the brain’s neuronal network computes a kind of a “hash code” represented by activation of an ensemble of neurons. Each neuron in the ensemble might fire in response to images of one person in a thousand, but the full ensemble would be almost unique for that particular individual–i.e. if you found two “Halle Berry” neurons, and surveyed enough images, you would find other people that each neuron responded to, but they would most likely be different people for those two neurons.
Nice idea trrll! and Larry, I was thinking something very similar… that face pattern neurons fire, and the cell they measured (and here I think of what eric posted) is more an “output” cell(s?) that triggers off into all the other places in the brain that perhaps store non face information like name, age, who they are currently dating, what makeup they use … you know, that kind of really useful stuff that we all store up often against our will (I know my brain is full of pop culture garbage whether I like it or not).
Meh, what do I know? But it is fun to think about it and try to come up with a solution
Endlessly fascinating.
But yeah, DrSteve I would love to know more about the study and I am fairly certain I wouldn’t understand the jargon associated with the actual study, so your translation for the less well studied in medicine would be great.
godkillzyou wrote:
“I’ve often wondered if “subconsciously” humans developed computers to work in the way they do because it mirrors the way our brains work…”
Classical computation, not at all. The two work entirely differently.
However, there’s a computer science application called “Artificial Neural Networks” (or ANN) that WAS based on the structure of biological neural networks. Just as the brain consists of an interconnected set of neurons, ANNs consist of an interconnected set of nodes. The pattern of connections, the strength of those connections, and the functions within the nodes to determine what to do as the result of stimulus from those connections ultimately determine the functionality of the ANN. An ANN is typically an adaptive system, which changes its own structure (eg: by changing the weights of certain connections) as the result of information that flows through the network.
An example application of ANNs is in pattern recognition. An ANN can be set up with no prior knowledge of any particular type of pattern. You then go through a “training” period where you provide input (eg: graphics, some of which represent a human face, others of which do not), let the ANN “decide” whether the pattern represents a face, and provide feedback as to whether it made the correct decision. As a result of this training, the ANN will adapt the ability to recognize graphics that represent a human face.
A Wikipedia article on ANNs can be found here:
http://en.wikipedia.org/wiki/Artificial_neural_network
Very interesting. Thanks for the input. (Pun?)
[...] Recent research by neuroscientist R. Quian Quiroga finds that human brains are assigned one neuron to remember a familiar face. Dr. Steven Novella discusses face-recognition and Single Neuron Neuroscience. [...]