Cognitive architecture
or
the neural signature of transparallel processing

The term cognitive architecture refers to computational models of not only resulting behavior but also structural properties of intelligent systems. These structural properties can be physical properties as well as more abstract properties implemented in physical systems such as computers and brains. There is no consensus about what these structural properties should be, and indeed, many different cognitive architectures have been proposed. These models differ, for instance, in whether they involve fixed or flexible architectures, in what forms of processing they allow, and in the extent to which they are based on a set of symbolic information-processing rules applied by one central processor or rely on emergent properties of many interacting processing units. Most models agree, however, that a cognitive architecture is a parameter-free blueprint for a system that acts like the human cognitive system as a whole.

Cognitive architectures differ from cognitive models and expert systems which usually (a) focus on particular competences such as language, concept learning, or problem solving, and (b) are judged solely by their resulting behavior. As indicated above, cognitive architectures are judged by their structural properties as well. Below, to address the question of what the structural properties of the human cognitive architecture in the brain might be, the focus is first on the visual hierarchy in the brain, and then on the intriguing phenomenon of neuronal synchronization which, in the visual hierarchy, seems to underlie perceptual organization.


The visual hierarchy

The top end of the visual hierarchy seems to involve a smooth transition into higher cognitive structures, while the bottom end can be said to be in the primary visual area V1 in the occipital lobe, which receives its main input from the lateral geniculate nucleus (LGN) (see next left-hand figure). In the LGN, a distinction can be made between retinal input entering the parvocellular pathway and retinal input entering the magnocellular pathway. Via V1 and higher visual areas, these pathways bifurcate into a ventral and a dorsal stream which seem to be dedicated to object perception and spatial perception, respectively (see next right-hand figure).

Retinal signals go, via the optic chiasm (OC) and the lateral geniculate nucleus (LGN), to the visual cortex; the OC arranges that the left-hand visual fields of both eyes are projected onto the right-hand cortex, and vice versa; in the LGN, retinal signals enter parvocellular and magnocellular paths, which perform something like a spatial frequency analysis.		In the visual cortex, the signals bifurcate into ventral and dorsal streams which are dedicated to object perception and spatial perception, respectively.

The neural network in the visual hierarchy is organized with 10-14 distinguishable hierarchical levels (with multiple distinguishable areas within each level), contains many short-range and long-range connections (both within and between levels), and it can be said to perform distributed hierarchical processing. Furthermore, in the visual hierarchy, the intertwined but functionally distinguishable subprocesses of feature encoding, feature binding, and feature selection seem to be mediated by feedforward (or ascending), horizontal (or lateral), and recurrent (or feedback, or reentrant, or descending) connections, respectively (see next left-hand figure). As indicated in next right-hand figure, these three subprocesses have also been implemented in SIT's transparallel processing model of perceptual organization (see also Smart processing and Hyperstrings).

(a) The three intertwined but functionally distinguishable subprocesses which, in neuroscience, are believed to take place in the visual hierarchy in the brain. (b) The three corresponding and also intertwined methods implemented in SIT's transparallel processing model of perceptual organization.

The three subprocesses are physically intertwined but, functionally, they can be characterized separately as follows:

• Feedforward connections seem responsible for a fast initial tuning, from lower to higher visual areas, to features to which the visual system is sensitive; during this so-called feedforward sweep, more structured information seems to be coded in higher visual areas (for a demo, see Time effect).

• Horizontal connections seem responsible for binding similar features within visual areas into feature constellations.

• Recurrent connections seem responsible for a top-down selection of different features from those feature constellations, and for the integration of these features into complete percepts.

There is controversy about the question of whether the selection of different features and their integration into complete percepts is controlled by endogenous, attention-driven, recurrent processing starting from beyond the visual hierarchy or by exogenous, stimulus-driven, recurrent processing within the visual hierarchy. This controversy seems moot, however. After all, the combination of feedforward and recurrent processing in the visual hierarchy might be analogous to the cascade formed by a fountain under increasing water pressure. That is, as the feedforward sweep progresses along ascending connections, each passed level in the visual hierarchy forms the starting point of integrative recurrent processing along descending connections. This yields a gradual buildup from partial percepts at lower levels in the hierarchy to complete percepts near its top end. This implies, on the one hand, that top-down attentional processes may intrude before a percept has completed, but on the other hand, that the perceptual organization process has already done much of its integrative work by then. To paraphrase Neisser (1967), before you can pick an apple from a tree, you first have to perceptually organize the scene to at least some degree.

Furthermore, in between feedforward and recurrent processing, there is the horizontal binding of similar features. This subprocess is a relatively underexposed topic in neuroscience, but it may well be the neuronal counterpart of the regularity extraction operations which, in representational approaches like SIT, are proposed to lead to structured mental representations. This subprocess seems to involve transient (i.e., input-dependent) neural assemblies which also have been implicated in the phenomenon of neuronal synchronization.


Neuronal synchronization

Neuronal synchronization is the phenomenon that neurons, in transient assemblies, temporarily synchronize their activity. Not to be confused with neuroplasticity which involves changes in connectivity, such assemblies are thought to arise when neurons shift their allegiance to different groups by altering connection strengths, which may also imply a shift in the specificity and function of neurons. Both theoretically and empirically, neuronal synchronization has been associated with a broad range of cognitive processes. Gamma-band synchronization (30-70 Hz), in particular, has been associated with visual processes such as those dealing with change detection, interocular rivalry, feature binding, Gestalt formation, and form discrimination.

The dynamics of neuronal synchronization (mostly in visual processing) are being studied by way of methods from dynamic systems theory. Furthermore, proposed ideas about the meaning of neuronal synchronization (also mostly in visual processing) are, for instance:

• that it underlies consciousness;
• that it is under the control of selective attention;
• that it is a marker that a steady state has been achieved;
• that its strength is an index of the salience of features;
• that more strongly synchronized assemblies in a visual area are locked on more easily by higher visual areas;
• that it binds different features into an integrated percept.

Indeed, neuronal synchronization may reflect a flexible and efficient mechanism subserving the representation of information, the regulation of the flow of information, and the storage and retrieval of information. All these ideas, however, are about cognitive factors associated with synchronization rather than about the nature of the underlying cognitive process itself. Therefore, instead of saying that synchronization mediates cognitive processes, it seems better to say that it is a manifestation of cognitive processing -- just as the bubbles in boiling water are a manifestation of the boiling process.

The foregoing does not make neuronal synchronization less interesting -- on the contrary, it raises the question of what form of processing it might be a manifestation. That is, it does not seem to reflect a simple form of parallel (distributed) processing. After all, basically, parallel processing is performed by different agents who simultaneously do different things. When these agents simultaneously do the same thing, however, they seem to enter another processing mode -- think of flash mobs or groups of singers going from cacophony to harmony. Indeed, assuming that neuronal synchronization underlies perceptual organization, which is characterized by a high combinatorial capacity and high speed, it must be a special form of processing that manifests itself by neuronal synchronization.

Notice in this respect that:

• in neuroscience, gamma-band synchronization is associated with binding of similar features rather than with binding of different features;
• SIT's model of perceptual organization puts forward a special form of processing, namely, transparallel processing by hyperstrings, where hyperstrings are distributed representations that can be said to bind similar features (see Hyperstrings).

These two things together suggest that gamma-band synchronization in the visual hierarchy in the brain might be a manifestation of transparallel processing of similar features gathered in hyperstring-like transient neural assemblies. As indicated next, this gives rise to a picture of flexible cognitive architecture implemented in the relatively rigid neural architecture of the brain.


From neurons to gnosons: a pluralist approach

The foregoing indicates that SIT's model of perceptual organization is neurally plausible in that it implements those three intertwined but functionally distinguishable subprocesses in the visual hierarchy. Furthermore, it suggests that the mechanism of transparallel processing by hyperstrings provides a computational explanation of synchronization in transient neural assemblies.

In SIT's model, the combination of feedforward feature encoding and horizontal feature binding yields an input-dependent tree of hyperstrings, that is, a hierarchical distributed representation which represents the output space for only the input at hand. This contrasts with standard connectionist modeling, which starts from a pre-fixed network that represents the output space for all possible inputs. The intertwined subprocess of feature selection and integration in SIT's model, however, is comparable to the method of selection by activation-spreading in standard connectionist modeling (see Inanimate PDP in Slimy, Hilly, and Pixy).

Hence, SIT's transparallel processing model of perceptual organization transcends traditional definitions of representational and connectionist approaches, in that it puts the representational idea that cognition relies on regularity extraction to get structured representations in a more dynamic perspective together with a more flexible version of the connectionist idea that cognition relies on activation spreading through a network. Its transparallel mechanism also relates plausibly to neuronal synchronization, so that it also honors the idea in dynamic-systems theory that cognition relies on dynamic changes in the brain's neural state. In this sense, SIT's model reflects a truly pluralist account of visual processing (see also Marr's levels and Metaphors of cognition).

Furthermore, the idea that cognition is a dynamic process of self-organization is not new, and the idea that those temporarily synchronized assemblies are the building blocks of cognition is not new either. What SIT's model adds, however, is the idea that the temporarily synchronized neural assemblies in the visual hierarchy can be seen as hyperstring-like cognitive information processors which, in a transparallel fashion, process similar features in the input at hand. Therefore, these assemblies of neurons may be called "gnosons", that is, fundamental particles of cognition, and they can be seen as the constituents of flexible self-organizing cognitive architecture in between the relatively rigid level of neurons and the still elusive level of consciousness.


For a demo on transparallel processing as such, see Pencil selection
For a formal account of transparallel processing by hyperstrings, see Proceedings of the National Academy of Sciences USA 2004
For extensive discussions and references on cognitive architecture, see Cognitive Processing 2012