User:Psyc4600/Group12

Grandmother cell refers to a set of hypothetical neurons that selectively respond to any complex, specific, and meaningful percept or stimulus. It is proposed that grandmother cells concatenate features of a stimulus, and thus differ from neurons in the primary visual cortex (V1), which only respond to particular features of a visual stimulus, such as lines, edges, angles, or movements. Furthermore, whereas neurons in the fusiform face area (FFA) can only discriminate between faces and non-face stimuli, grandmother cells can discriminate between more complex and specific stimuli, such as the face of Jennifer Aniston.

Origin

The term was coined in 1969 by a Massachusetts Institute of Technology (MIT) cognitive scientist named Jerry Lettvin, as an attempt to satirize the idea of having a single neuron represent a highly specific percept, such as one's grandmother. The parody featured a mother-obsessed character (based on the 1969 Philip Roth novel, Portnoy's Complaint) named Alexander Portnoy, who had all of his "mother cells" ablated by the scientist A. Akakhievitch. Akahievitch isolated a group of cells, which responded only to a specific mother, "whether animate or stuffed, seen from before or behind, upside down or on a diagonal or offered by caricature, photograph or abstraction." After the surgery, Portnoy's concept of his own mother was lost. Akakhievitch then went to study grandmother cells.

A very similar concept of grandmother cell had been developed a few years earlier, by Jerzy Konorski, who called these cells "gnostic" units. In 1890, American pyschologist William James developed the notion of "pontifical cells."

Functional specialization and organization in the brain

The medial temporal lobe

Hierarchical organization in processing of specific, meaningful percepts or stimuli, from basic features to the whole object

Studies indicate the possibility of grandmother cells located in the medial temporal lobe (MTL), an area which consists of hierarchical interconnected structures and areas, such as the amygdala, hippocampus, entorhinal cortex, parahippocampal cortex, and perirhinal cortex.

Certain parts of the MTL are thought to be functionally specialized (functional specialization[1]). For example, the hippocampus is implicated in object recognition.

The fusiform face area is an area thought to be specialized for face recognition. The FFA is located in the temporal lobe, on the lateral side of the mid-fusiform gyrus, an area shown to be highly activated when presented with face stimuli.

Theories of hierarchical organization

Like visual processing, processing of specific, meaningful percepts or stimuli, are thought to occur in a hierarchical fashion. At the bottom of the hierarchy are cells that encode for basic features such as colour, shape, or size. These cells integrate together to form more complex cells that code parts of objects. The integration of cells continues until you have reached the top of the hierarchy, which is recognition of a whole object. Grandmother cells are the cells highest in the hierarchy, as these neurons fire rapidly to an object in perception.

One of the biggest objections for grandmother cells is the fact that if we were to have a specific neuron that encodes for every recognizable part of our life, there would not be enough neurons in the brain to handle this job. Another problem is this method of encoding is extremely inefficient. There would need to be a vast number of undifferentiated cells kept in reserve to code for the new objects that one would be likely to meet in the future.

Support

Face selective neurons

Functional magnetic resonance imaging (fMRI) studies have shown domain specificity for faces. Visual neurons in the inferior temporal (IT) cortex of the monkey fire selectively to hands and faces and not fruit and monkey genitalia, providing evidence for between-category distinctions.

Individual specific recognition neurons

Whereas the aforementioned face selective neurons deal with between-category distinctions (such as faces and non-face stimuli), grandmother cells deal with within-category distinctions (such as between faces). One such study that addresses within-category distinctions between faces is the 2005 California Institute of Technology (Caltech) study by Quiroga et al.. When presented with pictures of celebrities, neurons were found to fire selectively to Jennifer Aniston. When showing the string "Halle Berry," neurons fired selectively, as though they would fire to a picture of Halle Berry, implying that the neurons are responding to abstract representations of objects, and not just visual representations of objects.

Characteristics of grandmother cells

Sparse versus distributed coding

Distributed coding scheme is referred to as a greater set of neurons that are activated in response to a stimulus such as a person or an object. A single neuron is involved in coding more than one thing. In other words, a neuron encodes little information about a specific object, therefore, many neurons are needed to be activated together to represent an object.

In sparse coding schemes fewer neurons are activated in response to a stimulus. Each individual neuron contributes more to a representation and so it encodes more information about an object. For example, to identify a telephone, about 3 or 4 neurons may be activated compared to distributed coding scheme where 10 or more neurons may be activated. Sparseness does have its advantage, especially for memory, because compact coding maximizes total storage capacity.

The main difference between distributed and sparse coding scheme is the number of neurons involved in coding a stimulus. Another distinguishing feature is that sparse representation is better suited for rapid learning without erasing previously stored knowledge. However, they are poor at generalization. On contrast, dense representation is better at generalization but when learning new things it erases previously stored knowledge.

At later processing stages in the brain’s object-representation pathway, neurons become increasingly selective for combinations of features, and the code becomes increasingly sparse. Grandmother cells are the theoretical limit of sparseness, where the representation of an object is reduced to a single neuron.

Invariant versus variant coding

If the neural representation is invariant, there will be identical activity patterns of the neurons when viewing an object from different viewpoints. If the neural representation is non-invariant, the activity pattern of a neural population will be different when viewing an object from different viewpoints.

Grandmother cells have invariant representation. They are independent of the viewpoint of an object, therefore, from any view of a person or object, the grandmother cell is activated.

Discussion

The existence of grandmother cells is an ongoing debate. Neurons in the medial temporal lobe (MTL) were found to be selectively activated by conjunctions of gender and facial expression, by pictures of particular categories of objects, such as animals, faces and houses, as well as the degree of novelty and familiarity of the images. A researcher, Quiroga, has been studying epileptic patients who had electrodes implanted in their MTL and showed them 100 pictures. These pictures included photos of landmarks, various objects, and celebrities. Single-unit recordings from the MTL have shown that there are highly selective cells that may respond strongly to different images of a single celebrity. In one patient, a single unit in the right anterior hippocampus has been shown to be selective of actor Steve Carell. In another patient, a single unit in the left posterior hippocampus was found to fire to pictures of the actress Jennifer Aniston . These cells in the MTL do have some similarities to grandmother cells but some argue that this interpretation is unlikely. First, it is implausible to identify one cell out of a few hundred million neurons in the MTL that would respond to one specific person or concept. Secondly, there is a high probability that the same neuron that responded to these actors would likely respond to other persons or objects if more pictures have been presented. Each cell probably responds to between 50 and 150 distinct individuals or objects which is suggestive of a sparse representation of information. As well, emotional responses to some pictures can influence the neuronal responses. In other words, if there is a strong meaning to an actor for one individual, the neuron may elicit a greater firing response.

The MTL is not necessary for visual recognition, it is involved in episodic/declarative memory. The response of the neurons in the MTL to the actors are more likely to be involved with the storage of new long-term memories related to him. The neurons might have a role in learning associations between abstract representations. Even though the cells in the MTL may seem like grandmother cells, it is far from it.