December 9, 2012

Cognits, Rhythms, & Memories


Moving away from the hippocampus to straightup networks, the other theme of this blog, I'm deciding to do a post of a general overview on cortical networks and memory in general. After reading Olaf Sporn's Networks of the Brain, which I recommend, and re-reading a review article on memory and 'the cognit' (large neocortical networks supporting this function), I feel more confident in my ability to read, understand, and soundly comment on it.  I'm sure this is a highly incomplete story for anyone that wants to get technical.  For that, I recommend the book below.

Originally, I read Fuster's book (1) on the distributed networks (for a class in undergrad where it was assigned) and, since then, the general idea revolving around cognition and distributed processing has captured my attention.  Broadly, the 'cognit', as a term coined by JM Fuster, is a representation of some knowledge that is located in the neocortex: there is a hierarchical order to cognits and are neural networks that are widely distributed throughout the cortex.  Importantly, these cognits are highly interactive with each other.  Fuster is mainly concerned with the neocortex and, as such, does not talk much about subcortical areas but does not dispute that these areas are, too, wired  into the fabric (they have to be!).

This idea builds on the a Hebbian synapse idea and cell assemblies (wiring of neurons 'together') but extends greatly upon it to include, most importantly, how cognition is generated.  It also extends on the fact that cognits can intertwine with other cognits and participate in several seemingly disparate processes.  The activation of this cognit that is serving a particular memory does indeed involve the original cortical areas (e.g. sensory) as well as far-reaching frontal areas. For me, the cognit and other distributed processing models make sense, especially in the light of "weighting" pathways depending on task/mental demands/properties and also makes perfect sense of how we can have knowledge constantly being integrated into memories and new associations that can activate memories.

One maybe doesn't see how this is different from other distributed network or associative models of the brain, but the idea is not necessarily that it's radically different in terms of networks in the brain. Rather, it's a new way of thinking of how these networks act and how these networks give rise to cognition (memory; mind).  While the book of the cognit goes through attention, memory, perception, language, and intelligence, many cognitive functions, very broadly, can be distilled down into memory, whereby the absence of this faculty would render most other cognitive properties (but not all), well, useless.

(read on for: modules, networks, rhythms, & the cognit

The review starts out by setting forward some basic tenets of cognitive neuroscience, largely modular zone vs network models.  Fuster then describes the cognits and how these two views coalesce together to form a unifying theory.

Modular Zones

electrical stimulation of the brain's surface
Simply put, these zones localize a function to a specific cortical region.  This means that a function, in this case memory, can be given an anatomical locus (module) and was  believed to be the case because for two large reasons: stimulation and lesion. Through electric stimulation of the associative cortex (those regions not directly sensory or motor cortices), very 'real' memories could be produced in patients.  This stimulation was done during neurosurgery (pioneered by Wilder Penfield) while the patient was awake, allowing the person to self report.  The second line of evidence for module zones came from discrete lesions of the neocortex, whereby a specific insult would kill a circumscribed area and produce memory deficits (in rodents it could be a very precise area; in humans these lesions were largely non-specific to a certain areas but some were more or less localized to a specific region).  In addition, through physiological recordings of sensory areas, regions would respond to only visual, auditory, etc stimuli.  From these studies, it did appear that function would be localized to certain regions.

Today, though we have more advanced methods, and usually to people outside the imaging field, fMRI is a certain kind of evidence for modular zones.  The problem with thinking this way is that fMRI only 'detects' the areas that contribute the most or use the most resources, not necessarily the 'supporting' areas that feed into that area - discussed more below.  Of course now we get way more complicated with "seeding" areas and looking at functional connectivity, which is awesome, but still leaves the fundamental problem.

Along these same lines, Fuster goes on to state that cognitive neuroscience today has a fascination with functional modules.  In physiology, he writes, that microscopic columns appear to be particularly responsible for sensory or motor function.  He gives the examples of the frontal eye fields (FEF) for "ocular motility" and area MT for visual motion.  This may be fine for specific lower-order functions like those mentioned but the danger comes in when one tries to extrapolate this principle into cognitive functions, such as perception, memory, or language (a la Broca's area) as a whole.

Main point: there do exist modules within the neocortex; however, a cognitive process, like memory, cannot be assigned one specific zone/region (re: Karl Lashley).

Network (memory) Models 
the modular zones mature!

Network models largely are rooted in such modular zones with respect to sensory or motor modules.  This is fine, as stated above. Classical network models build on these zones to form a hierarchical organization in sort of a modular way.  The cognit, too, builds on these modular zones specialized for sensory and motor processing, but the cognit produces increasingly, ever-larger networks that are highly flexible and bi-directional.  He states, and makes a good point that, "The cortex in its totality is a network is a truism.  What is far from a truism is the parceling of that gigantic network into the multiplicity of overlapping, interactive, and specialized memory networks that emerges from the recent studies..." (2). This model says that the neocortex is made of many cognits that are networks "dedicated to the representation and retrieval of individual memory and knowledge" (2).  Simple sensory stimuli activate these early sensory modules (phyletic memory) but, the more complex the stimuli get, the larger the (associative) network becomes.  More complex perceptual memories require larger and larger networks of representations -- he states at the base are the sensory qualia networks and, at the top, the more categorical, semantical, conceptual memories.  These involve larger networks that intertwine with others and dynamically interact in a flexible way.  He rightly states that this larger cognit theory does not deny highly associative areas (associative nodes) that, when lesioned, disrupts the cognit involved and can impair specifics of the memory or kinds.

The Unifying Structure of Perception/Memory & How We Build a Cognit

At a low level and the most basic, a [memory] network, in a fundamental way, is a collection of cell populations that are connected together.  Sounds like a cell assembly.  These populations, though, are sparsely distrusted throughout the neocortex, constituting a wide cortical network.  The populations that code for percepts and the populations that code for actions are all joined or associated together, based in root in relational coding.  This percept+action combo is what constitutes the memory cognit; remember, a cognit is a knowledge representation.  The association populations between the percept+action originally experienced activates together to give rise to this one memory cognit, which gives rise to the cognitive or mnemonic experience itself.

This is what I particularly like about this cognit idea: Higher up but still on the scale of supra-networks, a highly complex memory (he gives the example of an autobiographical memory) encompasses a large memory cognit with smaller ones below (hierarchy form) in distinct "noncontiguous" areas.  Those networks are the more concrete representations of the memory.  The large cognit is what brings all the small networks together to give a unified autobiographical, cognitive memory.
Notice the overlapping commonalities
between the perception & memory and note
the temporal coincidence. These features
give the percept & memory its "individuality"
Darker colors are meant to be 'closer' to you
while lighter colors are the ones
in the distance -- 3D effect, yeah..
source: LR Glover Designs - MS Paint
(based on picture in article (ref 2)) 
On this note, he states that the "individuality of memory" comes from the almost-infinite different interactions, and these are responsible for the temporal distinctiveness, the conceptual distinctiveness, the abstractive distinctiveness that encompasses the knowledge sets.  Higher up in scale, as I've emphasized  previously, memory cognits all interact and overlap with each other in a very high degree.  Because of this, any collection of inter-connected cells or nodes can take part in many memories, or knowledges, or items.  The structure of memory is largely associative in nature and these associative properties are what builds and inter-connects cognits and this is where he tries to unite Gestalt psychology and Connectionism. Gestalt theory proposes that perceptions are based on a relation, too, and these are 'isomorphic' relationship and the temporal coincidence of the external stimuli builds these relationships, thereby unifying percepts (a la Gestalt).  Bridging Gestalt and Connectionisim, broadly, through these temporal coincidences, networks are formed in a 3D hierarchical manner.  This is not like a pyramid hierarchy but rather a huge confusion when looked at closely but shows a very ordered hierarchy when taken together.

Rhythms and Sync'ing Cognits

Neural oscillations have been known for quite some time (almost 70 years; re: EEG recordings).  Not only do EEGs [large scale] detect rhythms but also local field potentials [small scale] display rhythmicity   These rhythms are categorized by the frequency of the wave cycle.  Different frequencies of oscillations have been related to psychological states and processes.  It was found that neuron clusters in the striate cortex (V1, visual cortex, Brodmann Area 17 - pick your naming system) fired in a synchronous way to visual stimuli.  It was thought that this was evidence for the binding of perception (the combining of various visual properties into a coherent whole) and eventually leading to perceptual constancy.  These synchronous oscillatory patterns appear to be properties from two sources: membrane changes in the cell that is biased by previous oscillations occurring around it and spike trains from dendritic and circuit factors of the network.  

Switching to how rhythms influence the totality of a network and generate large networks that sustain cognition: When humans are required to sustain attention and working memory for an extend period of time, a high degree of synchronicity and multiple rhythms (theta, gamma, beta, alpha) can be observed in associative cortices and these frequencies can modulate the other frequencies. We can take an example from top-down attention networks that feed into a memory cognit.  High frequency oscillations, namely those of the gamma and beta frequency, have been detected in the both frontal and posterior associative cortex.        These bands are particularly important in terms of working memory, which some see as the constant activation to keep the memory in attentional demand; meaning that it's thought that these oscillations reflect 're-entrant' activity into the activated (oscillating) networks and this directs or biases our attention to certain percepts or memories.

Integration & the Perception-Action Cycle

A cognit can represent semantic knowledge and autobiographical knowledge with similar network characteristics -- it is the smaller cognits that give to each their distinctive properties. On the whole, though, a 'memory' has the same overall structure.  Smaller cognits are modules to a degree (sensory, motor modules) and connecting nodes from associative areas allow many different, hierarchically organized cognits to interact and remain dynamic.  Fuster states that (2) separating out memories by content is "not very helpful to cognitive neuroscience of cortical memory." (re the autobiographical and semantic memory all involving a larger cognit that only defines itself by its smaller networks).   He states that a memory is a memory that should not be divided if you want to think about the neuronal and network correlates of the mnemonic structure and function.  He puts this, I think, elegantly when he says that memory should  not be thought about in terms of 'systems of memory' but rather 'memory of systems'.  Simply, cognitive networks interact with sensory and motor networks to give rise to behavior through performing exactly what these smaller networks were formed for (through phyletic memories) and support goal-directed behavior through this interaction of frontal and posterior cognits (as described above & blogged on before).  In hierarchical format, more complex actions and more complex resources (executive memories) needed to reach a complex goal, the larger this cognit for 'GOAL' becomes.  This entire 'GOAL' cognit integrates "subjacent" cortical cognits that represent the easier, short-term goal-directed tasks (ex: lower-order: move hand 30 degrees upward vs. higher-order motor plan: integration of all movements needed to grasp object and drink juice).  

Perceptual memories are largely located in the posterior regions of the cortex while the more complex, executive memories are located in the anterior regions of the cortex.  These regions are interactive and integrate simple and complex cognits.  The working of these regions and memories gives rise to the Perception-Action cycle, which operates at all levels of the nervous system (spinal cord to the neocortex).  Once this cycle is initiated for some goal, serial and parallel processes occur to integrate other cognits or networks of the hierarchy.  Perceptions feed into actions as well as the actions feed into perceptions.      Habit memories, those that are well learned or instinctual largely do not depend on the neocortex but rather the subcortical cognits.  In these habit situations, the outer cortex only becomes engaged when there is an abrupt change, such as ambiguity encountered that is not pre-planned in the cycle.  From this, we go towards higher-order perception-action cycle memories.  Namely, those of executive memories are pre-planning.  Frontal networks are engaged in the mental planning of serial movements and these executive frontal regions seem to be a memory for the future (*raises eyebrow).  Similarly, those involved with this pre-planning are also involved in coordinating behaviors, beginning with witnessing the percept in space to the set of instructions to act upon it.  The viewing of animate object activates posterior regions of the cortex and those of tools (with a pre-planned motor use) seem to activate pre-motor cortex.


Memories constitute a widely distribubted, interactive, hierarchical network of cognits across the neocortex.  This encompasses posterior and anterior regions for a single memory.  It is the associative regions of the cortex that join the cognits together.  Cognits can integrate new information and can be expanded by experiences.      


1 comment: