Distributedness

Neuroanatomical appreciation of synaptic structure gave credence to connectionist theory, and permitted Hebb to extend the theory of synaptically regulated, discretely assembled neural nets to memory and behavior in humans. However, the only experimental data came from 30 years of work by Karl Lashley (1929, 1950) who had shown that memory and perception appear to be distributed throughout the brain. Lashley's quest was the site of representation— the "engram"—of information, memory and learned behavior in the brains of laboratory animals.

Lashley's experiments in search of the engram consisted of ablation of specific parts of animal brains with careful testing for retention of habits learned before the ablation, ability to relearn, and ability to learn new tasks. The overall conclusion was that memory function is disturbed in proportion to the amount of cortex destroyed, irrespective of which part of the cortex had been removed. As far as learning, Lashley felt that all parts of the cortex were equipotent except for the specific sensory receiving areas involved in that particular learning modality. Lashley's findings were a disappointment to connectionists and engendered a bleak, hopeless outlook. In his famous life's work, In Search of the Engram, Lashley (1950) wrote that after 30 years of searching for the location of the engram in the brain he was convinced that "learning is just not possible. Nevertheless, in spite of all the evidence against it, learning does sometimes occur." Lashley rallied from despair to propose an alternative to localized storage of memory traces attached to spatially fixed reflex pathways. Agreeing that recall involves reactivating a previous pattern of neuronal excitation, he insisted that the pattern representing any one memory could be evoked not just in one specific set of neurons, but in many sets, in many places, and perhaps anywhere in regions of the brain having to do with memory function. This distribution of information in memory was explained by Lashley as excitatory patterns in cortex related to the spread of interference patterns on the surface of a liquid disturbed at several points at once. (His theory of interference patterns later inspired comparisons between the brain, storage of information and laser holography).

Lashley (1929):

Nerve impulses are transmitted ... through definite intercellular connections, yet all behavior seems to be determined by masses of excitation ... within general fields of activity without regard to particular nerve cells. It is the pattern and not the element that counts.

Hebb (1949) commented:

Lashley has concluded that a learned discrimination is not based on the excitation of any particular nerve cells. It is supposed to be determined solely by the pattern or shape of the sensory excitation ... this suggests that the 'nmemonic trace', the neural change that is induced by experience and constitutes memory, is not a change of structure ... [it] is a lasting pattern of reverberatory activity without fixed locus like some cloud formation or an eddy in a mill pond.

"Engram" was originally described as the brain's "nmemonic" trace of an elementary idea as if it were the atom of mental content. An earlier proposal was that one engram was stored in the firing pattern of one cortical neuron: a "psychon" within a "grandfather" neuron at the apex of a neuronal hierarchy. Pulse logic saw "psychons," (or engrams) in the discharge patterns of neurons and connectionist neural nets equated engrams with specific circuits of synaptically connected neurons. Activity within those circuits was thought to represent consciousness and memory that would occupy specific locations within the brain. Memory was viewed as libraries, filing cabinets, digital computers, and junk boxes in which information was stored in particular places and retrieval involved finding where it was stored. Lashley's experiments suggested that information is not stored anywhere in particular, but rather is stored everywhere; memory was distributed. One prevalent interpretation was that information was stored in the relationships among units and that each unit participated in the encoding of many, many memories. Information could thus be distributed over large spatial areas. With distributed memory, individual traces from within a complex of traces can be found much like the way filters can extract individual frequency components from complex acoustic waveforms. Filters are able to detect the presence of specific frequencies even when they are completely intertwined with others. Consequently, a filtered, distributed memory system can operate as a storage and retrieval device. If memories are not independent of one another, the storage of one memory can affect another. Previously stored information tends to evoke an original pattern of activity even though the inputs to the system may differ in many details. This description is similar to the hologram concept in which coherent reference waves are necessary to retrieve information from an interference pattern.

Lashley (1950):

It is not possible to demonstrate the isolated localization of a memory trace anywhere within the nervous system. Limited regions may be essential for learning or retention of a particular activity, but within such regions the parts are functionally equivalent. The engram is represented throughout the area. All of the cells of the brain must be in almost constant activity either firing or actively inhibited. Every instance of recall records the activity of literally millions of neurons. The same neurons which retain the memory traces of one experience must also participate in countless other activities.

Recall involves the "synergic" action or some sort of resonance among a very large number of neurons ...

Hebb's view of engram representation was a closed loop of neurons firing in a confined region. Lashley's studies led him to suggest an open, parallel, distributed network covering wide regions of brain. Hebb's linkage of learning and synaptic efficacy transcended this conflict because it related to both concepts, which may also operate within cytoskeletal networks.

Figure 4.3: Brain memory bank based on "conventional" notions of dendritic spine synapses. Input lines (B1-3) are axons which branch and synapse on dendritic spines (E) connected to output lines (C). "File dump lines" (A, d1-3) direct spine activities to "copy" or "dump." Such a configuration allows each spine to hold about 3 bits of information, remarkably low when compared to the capacities of biomolecules such as DNA or microtubules. Cytoskeletal activities within dendritic spines may contribute. By Paul Jablonka.

Figure 4.3: Brain memory bank based on "conventional" notions of dendritic spine synapses. Input lines (B1-3) are axons which branch and synapse on dendritic spines (E) connected to output lines (C). "File dump lines" (A, d1-3) direct spine activities to "copy" or "dump." Such a configuration allows each spine to hold about 3 bits of information, remarkably low when compared to the capacities of biomolecules such as DNA or microtubules. Cytoskeletal activities within dendritic spines may contribute. By Paul Jablonka.

0 0

Post a comment