Working Memory

Working Memory – Many researchers have made valuable contributions to the curated research literature underpinning the CFDI Therapeutic Framework. As is the nature of scientific research, though, empirical progress is often made at the cost of theoretical understanding. Useful frameworks generate research questions. Answers to those questions typically drive refinement to the conceptual framework or cause them to be discarded in favor of concepts with better empirical support. Such is the current state of understanding among neurocognitive and physiopsychological researchers regarding working memory.

Working memory has a unique place among the collection of primary cognitive functions within the CFDT therapeutic model. Whereas empirical evidence indicates that attention and memory arise from internal cognitive processes supported by neural network activity, the same cannot be said of working memory. Working memory exists only as observable external behavior. It does not map back to independent internal cognitive processes or underlying neural network activity. Instead, it arises as a result of the attentional system interacting with memory by way of coupled neural oscillations.

Working Memory as the Intersection of Attention & Memory

Memory information deemed task-relevant by attention is transiently encoded as gamma frequency bursts held within a coupled theta or beta frequency. In this form, the information may be acted upon by any number of top-down processes or influenced by one or more bottom-up processes before generating neuroplastic remodeling as learning or new information.

Working Memory and Attention

There is increasing neuroimaging evidence that attention and working memory rely upon common neural substrates. These substrates independently incorporate each of the five physical senses. That is, there is a working memory/attention substrate for vision, one for hearing, and so forth. It is reasonable to posit that in addition to neural substrates associated with each of the five senses, there is another similar internal attention/working memory substrate associated with the medial temporal lobe (MTL) memory system. The MTL is responsible for declarative / explicit memories including somatic and emotional (i.e., non-sensory) memories.

The attentional system may be further subdivided (within each of the neural substrates) as being composed of three subnetworks – alerting, orienting, and executive control. Within this framework, the alerting network provides responsiveness to sensory stimuli.  Inasmuch as sensory stimuli are first pre-processed through the corresponding sensory cortex (or the MLT for non-sensory stimuli) and thereby generates so-called sensory memories, the CFDI therapeutic model conceptualizes all information deemed task-relevant by the alerting system as fist passing through and receiving initial encoding by the various sensory or MTL memory systems.

The determination of task-relevance is a dual process handled by the alerting system and the executive control system. With regard to the former, task-relevance is determined by having a sufficiently high level of perceptual load such that distracting stimuli are “squeezed out” of the limited capacity of working memory/attention. An example of this phenomenon is driving in fair conditions vs. adverse conditions. A primary, task-relevant stimuli source for driving is the forward-facing visual field. In fair conditions, when incoming visual stimuli are highly predictable, an experienced driver has enough attentional/working memory resources to attend to the visual field and to carry on conversations with passengers safely. In adverse conditions, however, when incoming visual stimuli are highly unpredictable, the driver may request that passengers keep their discussion at a low volume and exclude attempting to engage the driver. The task of ingesting and processing the forward-facing visual field is necessarily consuming more resources and squeezing out room for processing distractor stimuli such as conversations with passengers.

Once information is deemed task-relevant, the orienting network selects a subset of the stimuli for priority processing. Such processing may include suppression of distractor stimuli (as in the driving example), priming to orient toward similar stimuli, and boosting of neural representations through the addition of neuromodulators known to be involved in recruiting attentional resources and strengthening memory traces.

The executive network processes post-sensory representations and resolves conflicting demands for access to limited resources. The executive network is also active in allowing simultaneous storage and processing of information. In this sense, working memory and the executive control attentional network are synonymous with both the functional and neural network levels.

Working Memory and Memory

Stored information and sensory stimuli are introduced to the neural landscape as recalled sequences of stored patterns or as encoded sensory memory. It is in this memory-encoded format that the attentional system can select task-relevant information or stimuli.

The capacity of working memory, as discussed in the next section, is generally limited to seven +/- two items. Relevant information required beyond this capacity can be temporarily off-loaded from working memory to short-term or intermediate-term memory as dendritic spinogenesis.

Working Memory as Neural Oscillations

Working memory stores and carries items as discrete bursts of gamma oscillations, rather than by neuronal activity. Maintaining multiple items and performing computation on carried items requires nesting gamma bursts within theta-band carrier frequencies.

Interestingly, the transient phase-amplitude coupling of gamma and theta frequencies gives rise to the so-called “magic number,” denoting working memory’s limited capacity. In general, working memory’s capacity is limited to seven items, +/- two. The actual number of items that can be transitively stored is dependent upon the nature of the item (for gamma burst encoding purposes) and whether the items are novel or familiar. Familiarity with the items allows for “chunking” of information into representational blocks, with storage for the individual items handled by existing memory. Each item – from novel information to familiar representations of “chunked” memories – is represented in working memory as a gamma frequency burst. Each gamma burst then occupies “one space” in the carrier wave cycle, which oscillates at either a theta band or a beta band frequency. Generally, only seven such bursts, plus or minus two, can fit within a theta-band carrier wave cycle – thereby giving rise to the “magic number” of working memory’s limited capacity.

 

Therapeutic Delivery for Working Memory

In many ways, the therapeutic goals for developing working memory will closely follow the goals for improving attention and memory. However, since working memory exists as the interaction between these two functions, therapy will focus on the proficiency with which these interactions take place. Note, for our purposes, we define proficiency as a dynamic exchange between speed and accuracy. That is, when either speed or accuracy is the primary concern, the other component may take a backseat while preserving the same level of proficiency.  

Increased Proficiency Transferring to Working Memory

For items to be stored in working memory, they must first be transferred from the memory system, either as sensory memories or recalled encodings stored on dendritic spines.

• The first step is to distinguish relevant stimuli or information from the myriad of non-relevant or distracting stimuli and information. The attentional alerting network handles the bottom-up evaluation of relevant stimuli; the attentional executive control network initiates the top-down effortful assessment of what is relevant. Therapists should observe client behavior for the ability to select items on command or to notice a subset of items matching given criteria. For example, if all 52 playing cards are placed face-up in an orderly fashion before the client, the client should be able to distinguish and quickly select from all the cards only those matching the specified criteria, such as all red cards with a value of five or less.

• Transference to working memory begins once relevance is recognized. Increasing the proficiency of this process is the bailiwick of the attentional orienting network. Orienting allows for priority processing of the relevant set of information, or a subset thereof. Doing so further pushes out distracting stimuli and will temporarily recruit additional neural computational and storage resources to handle the selected set or subset of information. It is posited that herein are marshaled the ability to generate theta-gamma coupled neural oscillations and stimulate spinogenesis for short-term and intermediate-term memory.

Increased Proficiency Carrying Items in Working Memory

One of the critical distinctions in the CFDI Therapeutic Framework is acknowledging that an individual’s capacity to hold several items in working memory is a matter of physics, not neuro networks, cognitive processes, or other neuroplastically changeable facets. Thus, the therapeutic goal is not to attempt to increase the working memory capacity. That number is fixed by the theta-band wavelength and the size of the nested gamma bursts. Instead, the goal will be to increase proficiency carrying – that is, accurately maintaining – gamma-encoded representations.

• The length of time in which a non-rehearsed item can remain encoded in working memory is on the order of 1-2 seconds if the item is attentionally salient (i.e., deemed task-relevant and produced a gamma burst with sufficient amplitude), less otherwise. Attentional saliency can be augmented through a top-down effort from the executive control network. Thus, increased proficiency in working memory carrying capacity can be developed as a consequence of improving the client’s ability to sustain executive network effort. A standard, non-therapeutic approach to this is the phrase, “Hey, pay attention,” which is a demand for top-down effort. A better approach, of course, is to exercise and strengthen the executive control network in conjunction with mindfulness (a higher-level cognitive process) such that the individual is capable of self-generating the reminder to “pay attention” and able to muster the resources to do so.

• If the task requires maintaining items in working memory longer than 1-2 seconds, the gamma burst representations will need to be refreshed. Refreshing may be accomplished by the repeated rehearsal of the carried items, such as repeating a grocery list to yourself on your way into the store. However, repetition is subject to error introduction and distractor interference. Increased proficiency can be gained by developing the client’s ability to alert to and resample the contents of working memory. In doing so, the same processes apply as when first transferring information to working memory, typically involving the effortful top-down processes of the executive control network.

• As an alternative to effortful refreshing, the individual can off-load item representations into short-term or intermediate-term memory. Doing so will allow items to be stored for up to eight seconds with near-immediate access. The trade-off, of course, is that off-loaded items are not available for computations or other manipulations. The process of off-loading will necessarily require spinogenesis. If sufficient attentional saliency exists, off-loaded items may enter long-term memory with resultant synaptogenesis. When this occurs, we say the item has been learned, and a long-term change in behavior can be observed.

Increased Manipulation Proficiency

Only items represented as gamma frequency bursts can be utilized in computations or otherwise manipulated to generate a new set of encoding. For example, most individuals who experience CFDT have a long-term encoding of the meaning of 7 and 8. Unless the product of these numbers is also encoded and readily recalled, the only way to output their product is to compute it while holding the items in working memory.

Based on the preceding discussion regarding the capacity of working memory and the duration of items carried, it should be immediately apparent that moving beyond the most straightforward computations will require proficiency with transference and carrying. However, manipulation proficiency also requires that an individual prioritizes and organizes transference and carrying activities in a task-relevant, goal-oriented manner. The capacity to do so is almost exclusively the bailiwick of the executive control attentional network. Developing this network can either be accomplished through activities designed to target it or though activity loading.

 

Other models of Working Memory

Since the research literature draws from multiple concepts of working memory, it may be useful to provide a brief overview of the models to which more common references are made.

Multicomponent Model

The multicomponent model of working memory was the leading view from 1974, when it was first presented by Baddeley and Hitch, until the early 2000s. Within this model, working memory is conceived of as having four primary components, each of which has capacities for information storage and manipulation, and which can communicate with one another and with long-term memory:

• Phonological Loop – A “relatively modular system” comprised of transitory storage and a means of refreshing information in storage through vocal or subvocal rehearsal (i.e., repeating a grocery list to yourself on the way to the store). It is believed that the phonological loop supports the acquisition of language through the temporary storage of new words until they can be stored in long-term phonological memory. The phonological loop is involved in receiving and processing auditory information, such as a list of digits. It also plays a role in reasoning, hence the popular pastime of talking to oneself while working out a difficult problem.

• Visuospatial Sketchpad – The purpose of this component of working memory is to represent, maintain, and allow for an understanding of information that can be represented visually and/or primarily, and to do so in a way that persists across an individual’s irregular pattern of eye movements as he or she scans the visual world. For example, whereas the phonological loop is used directly to receive and process verbal directions to a location, the visuospatial sketchpad would be employed in translating the verbal description into a mental map. Alternatively, the sketchpad would be used in interpreting a printed map relative to observed landmarks and topographical reliefs. As with verbal rehearsing to keep items in the phonological loop fresh, items stored in the visuospatial sketchpad may be refreshed by “covert motor performance that serves to reactivate the memory traces residing in sensory stores,” such as eye movements. Thus, by rehearsing the eye movements, one might need to observe a physical object; the memory of the object is refreshed in the sketchpad.

• Episodic Buffer – A passive component of working memory that allows “binding” of small chunks of multidimensional information – or episodes – from long-term memory, sensory inputs, and other working memory components. The episodic buffer seems to play an important role in consciousness. However, there is still ongoing academic discussion as to the operation, purpose, nature, and even the existence of this component.

• Central Executive – Within the multicomponent model, the central executive is believed to be an attentional control system. While this component is considered to have limited processing capacity, it is crucial in overall control of an individual’s actions. Within this model the central executive function may be divided into two attentional subsystems: an automatic system responsible for handling well-learned or habitual behaviors, and the so-called Supervisory Attentional System which freely draws from long-term memory and other working memory components to postulate on possible outcomes to a given situation, and then chooses the solution deemed most desirable.

Global Workspace Theory

The Global Workspace Theory postulates that the brain is comprised of a highly interconnected network of specialized processes, what Baars denoted as “a brainweb.” Coordination and control of this web take place through an individual’s consciousness – via an architecture which is like, but not necessarily identical to the multicomponent model’s episodic buffer. This theory allows for multiple, distinct, and separately acting conscious activities – e.g., auditory and visual consciousness, each capable of being willfully activated, each located in a different part of the brain. Similar to the central executive component of the multicomponent model, an individual’s activation of conscious functions requires the utilization of the selective attention system under control of both the frontal executive cortex and automated centers such as the amygdala.

Recently functional connectivity analyses have lent support to the Global Workplace Theory.

Bayesian Probabilistic Inference

Also termed “The Bayesian Brain,” this model builds off the Global Workspace Theory to describe how unconscious backstage processes can guide decision-making processes under conditions of uncertainty. Three main principles are worth noting:

• Activation of the prefrontal cortex and hippocampus neural networks allow the individual to determine, albeit without conscious awareness, a prediction error and to infer an evaluation of current experience based on prior experiences.

• Neural systems within the prefrontal cortex, basal ganglia, and insular cortex derive or interpret familiarity with a given stimulus based on the stimuli’s salience, and thereby merge prior experiences with current beliefs about that stimuli

• Activation of the dorsolateral prefrontal cortex, visuospatial, and language centers allows processing of the exposure frequency to specific events for the individual to develop a belief system about the stimuli, leading to making predictions of how those stimuli will behave and how the individual should react should the context become uncertain.

Within this model it is expected that there will be a certain level of epistemic foraging for information – that is, there exists some tendency within the individual to sample both intrinsic and extrinsic stimuli until the individual is able to ascertain which of the available predictions about an uncertain future is best, given prior beliefs and experiences. This concept becomes significant when developing and delivering therapeutic procedures to stimulate neuroplastic processes to improve mental health.