Patterns: Information representation, embedding and recognition

Real-time state changes in the behavioral temporal spectrum, and representational grain

The temporal information content of spatiotemporal patterns encoded in spatial form is not discarded nor ignored. The agent is interactive in the world, coupled to natural physics. In it’s environment it’s system must represent behavioral patterns in natural time scale orders, thus requiring tracing and regeneration of relatively slow pattern evolution. E.g. patterns up to an upper resolution limit in the order of 10^3 Hz, lower resolution limit in the order of 10^-2 Hz, and up to 10^2 Sec in duration typically apply in the world of macroscopic biological systems.

Activation of encoded patterns underlies potential recognition and facility to reconstitute (or compound in novel ways by applying some learnt or intrinsic operator function), thereby enacting pattern composition and evolution in sensorimotor scale real-time. The decoding recognition function involves in-time detection and potential completion or reconstitution (gestalt) – a reverse transformation – from the spatially embedded form into the spatiotemporal enactment.

Since an agent’s computationally manageable representational grain is not infinitesimal (spatially and temporally), since both it’s representational and connective capacity are comprised of a limited set of discrete resources, the agent optimizes it’s adaptability by employing selectivity and compression. The system represents salient pattern factors as compressed invariants, and salient relations as the enact-able affinities between these. A range of, or several primary, characteristic resolutions and spans applying to pattern factor representations are likely to be employed, to optimize overall representational efficiency and robustness (i.e. a rough spatial and temporal frequency band separation/interaction mechanism).

Pattern component representations making up an agent’s instantaneous egocentric state span immediate past to immediate future periods. Ignoring major performance-directing influences by attentional effects, the system’s predictive horizon is a function of the effective span of pattern factors having some arbitrarily demarcated representation in the system: Pattern components of longer duration imply slowly evolving activity, that is grounded  in some form of explicit invariant representation. The extent of the projection into the future that the system is capable of, before attentional/intentional/motivational effects come into play, is one of it’s key performance characteristics. See The low frequency challenge and spatial encoding, below.

The system anticipates it’s next state just as that next state is developing, by utilizing all the information that is available to it as enactments of it’s statistical records of state evolution from near past states – to maximize correct world matching instantaneous activity. Therefore notions of, or distinctions between, “bottom up” and “top down” functions of information processing and pattern consolidation become blurred, intertwined and superimposed. The ideal instantaneous activity state would be grounded in a combination of the correct set of sensory matched (minimal unmatched residual) differentiated, embedded and enacted pattern representations, given the system’s spectral bounds (resolution and signal to noise limits) and combinatorial capacity.

A consensus supported by unequivocal experimental evidence supports the notion that neural correlates of conceptual thought and un-acted intent unfold in real time, and share elements of representational units and connections with behavioral/motor expressions (e.g. activity patterns of mirror neurons; motor actions as sensory expectations). Functionally even the notion of ‘anticipation’, as used above, can be interpreted as a generalization of the paradigmatically used term ‘preafference’.

Therefore an aspect of the system’s function is that of a real-time pattern-factor-representation combination based universal-approximator. As such it is obviously associational and it generates mental models (critique relating to shortfalls of associational theory of mind is dismissed as being inapplicable to but one principle aspect of the functioning of an elaborate machine).

Obviously the ideas being presented constitute a very rough simplification as they are intended to identify and describe principles of operation rather than dynamics in real behaving agents. For example in a behaving agent there may be evolved module functions distinctly supporting anticipation-bias vs actualization, putative correlates of which could be the empirically found hypocampal compressed-time sequences.

The particular model presented in this work emphasizes ‘level free’ temporal symmetry about the ever shifting present, that is expressed through complementing in-time isomorphic transforms between patterns and distinct stable representations.
I submit below that distinct spatially-stable and temporally-extended representation activation, throughout the encoded/recognition duration of any particular compressed pattern factor, is a defining characteristic of the ‘reciprocality’ of such isomorphic transforms. Temporal ‘smearing’, as exhibited by the particular form of encoding/embedding suggested and elaborated below, is a natural solution answering the need to factorize, segregate, relate sub-patterns and predict combined pattern evolution in a contextually reactive interacting system, aligned on a biologically scaled ‘now’. Furthermore it is proposed below, and in following sections, that such reciprocal isomorphic encoding, along with the related recognition/bias/enactment temporal dynamic, is suggestive of conceptual binding on the computational level, and may be required for intentional cognitive function.

But the utility of the model may extend beyond the neural-like computational domain. Arguments are presented below to support the notion that synchronicity is principle to perceptive and functional state. It is speculated further that physical coherence could be a hallmark of consciousness. Thus the theory emphasizes the ‘resonant’ characteristics of the model it describes, in some contrast to other principally ‘generative’ models. However, the model differs from other proposed resonant models in it’s conception of absolute coherence and resonance as principally underlying the formation of ‘I’ and order of scale spanning relations that bind the phenomenon of consciousness, versus emphasis on particulars of computational implementation e.g. in various flavors of ART (Adaptive Resonance Theory, Grossberg).

Projections, Isomorphic iconic encoding and virtual world model dimensions

Avoidance of useful information loss must be an evolutionary optimized capacity and a functional optimization target for an embodied behaving system in it’s environment. Isomorphic encoding conserves information (by definition) and is inherently verifiable. The combinatorial explosion of possible interactions between representation factors, as well as the coupling of representation evolution to myriad external influences, should favor compressed pattern factor encoding that supports universal re-combination i.e. optimizes resonant iconic pattern tracing capability. An entity’s ‘operating’ world model could thus be enacted, in part, by the reconstitution or tracing, through isomorphic compressed-encoding inversion, of iconic representations of perimeter signal patterns.

To an observer bounded by a perimeter and interacting with it’s environment (exclusively through it’s perimeter), an ‘internally’ generated iconic projection of a pattern upon the sensory perimeter (e.g. enacted pattern completion) may be indistinguishable from an externally stimulated signal pattern.

Therefore an agent’s world model may be comprised in part of sensed or re-composed/enacted patterns of environmental projections upon it’s physical sensory perimeter, equivalent, from within the bounds of the agent’s embodiment, to those “invariants” “which would project” such patterns upon the perimeter (discussed further in following sections).

Generally, the spatial span of a virtual world model that is iconically defined by the tracing of projections upon an agent’s bounding perimeter extends naturally to encompass the agent’s environment. Composite iconic representations can entangle patterns perceived from different perspectives and by multiple sensory modalities. ‘Multi-angle’ and multi-modal contributions to the perceived world model assist in disambiguation of ‘that which would project’. e.g. First person spatial experience is imbued with a sense of passage in time implied by movements required to both (literally) visually scan and (virtually) span distances (regardless of how the flow of time is represented and the associated phenomenality of duration comes about, though the latter seems to be a second order association, and the composite notion of action in time seems integral to the notion of distance in ego-centric peri-personal space).

The notion of reconstitutable explicit iconic representation is reminiscent of Gibsonian ideas concerning ‘direct representation’, but not at all with regards to exclusively bottom-up processing directionality. It is when this notion is considered as a functional principle of the computational aspect pertaining to theory of conscious agents that it attains deeper significance: As a modulator of phenomenality it supports and leads-in to ideas concerning consciousness manifesting in embodied agents, in the world.

A proposed modulatory relation between the computational process and phenomenality could be partly consistent with philosophical dual aspect theories, such as Reflexive Monism (Max Velmans; an appealing aspect of Velmans’s  view of consciousness in agents is the proposed dual aspect dissociation/relation between the unconscious-computational and conscious-phenomenal aspects; to be discussed further in following sections). However additional principles need to be introduced in an attempt to narrow the computational-phenomenal explanatory gap, to bore deeper into the nature of the seeming inclusion of observed in observer (binding etc.).

As neural like system processes involve an array of evolutionary advantageous computational supporting functions, obviously the above charts an extremely simplified schematic of a representational principle. For example visual sensory processing is entangled with motor actions and world model stabilization adjustments, yet that extra processing does not conflict with the notion of iconic representation. All of the above does not preclude an expression of ‘internally’ generated domain or task specific representation utilities: Anchors, abstractions, algorithms or mechanisms etc. (‘scaffolding’) possibly as virtual environmental entities providing reference and enhancing recognition and predictive capacities. e.g. Normalising and amplifying attentional mechanisms; Radial egocentric or allocentric spatial distance scales (as proposed to have been found in aspects of activity in the Entorhinal-Hippocampal regions, which by conventional interpretations, support an agent’s spatial location and path references). Neither does the principle of iconic representation preclude the utilization of various filtering and lossy compression techniques. The employment of such utilities must harmoniously support the embedding, recognition and reconstitution of perimeter bound pattern projections.

Unified iconic encoding-decoding (recognition or enactment)

The agent’s system must be able to learn and embed the Bayesian associations relating a rich set of self organizing multiple orders of spatiotemporal sub-pattern representations (‘pattern factors’), as these occur from time to time on it’s perimeter, or can be derived through operations upon those signals projecting on it’s perimeter. It factors the ever evolving projection signals into several orders of spatially embedded elements and cross association affinities (i.e. both of internally cohesive and contextual-composition relations, see below) – and it is able to use these factors and associations to effect the inverse process of coalescing and attracting, to re-enact spatiotemporal composites. This refers to essentially the same idea as presented by Tononi in his discussion of diversity and integration. Embedded representations may also include associations with modulators and logical function operators: These would be associations with pattern factors coupled to control of operations, and with pattern factors coupled to utilities that participate in, or support, such factoring/composition (be these hardwired or learnt, widely applied or narrowly specialized).

In-time capabilities would be elegantly realized if factorization, recognition and composition were executed as aspects of an isomorphic translation process: A process involving in-time accommodation or re-enactment of a stimulus pattern, through activation of it’s factors and the affinities relating them, just as these are being recognized while the composite pattern evolves. The isomorphic mechanism of encoding/enactment synchronously translates between isotropic configurations of activity and respective patterned activity by way of reciprocal embedded associations. Both enactment and recognition are essentially effected as bi-faceted filtered recruitment (see below and next section).

Prominent predictive coding models (e.g. Friston) are based on top-down vs bottom-up information stream comparators, and comparison requires isomorphic formulation. But the focus there tends to emphasize prediction vs error propagation and the generative capacities of these models do not necessarily imply capacity for iconic reconstitution, which would facilitate accurate factorization. The in-time state evolution proposed here does not stress ‘top-down’ notions over ‘bottom-up’ integration. Furthermore, communication paths through structured hierarchies of resolution levels might be lossy. The approach outlined here can be considered a ‘resonant’ variant extending ‘predictive’ models, and multi-scale combined rather than strictly hierarchical (while it differs from Adaptive Resonance theories in perspective and emphasis). Also worthy of note is that any ‘in-time’ system possessing direct or indirect self-sensing capacity would also exhibit recursive behavioral characteristics rendering literally coherent states.

An efficient realization could conceivably be locally self-organizing (in respect of encoding) and auto-validating. Since pattern encoding by factorization is likely to be gradual – a multi-’step’/multi-unit, or modular, ordered progression of transformations (until all information is encoded) – local iconic factoring-enactment could be applied throughout – at each and every ‘step’ in the progression of transformations, and scale free across orders of processing.

Continuity, contemporaneous state evolution

One perspective on the system’s operational goal is that it attempts to anticipate the content of its (conscious) world model in the next moments, inclusive of future input. In an environment exhibiting regularities, the maximization of patterned representation (extraction of state evolution information) co-manifests as a maximization of continuity in state evolution (where continuous implies isomorphic expression of embedded information, enact-able over and in time). i.e. continuity is effected through predictive capacities; prediction is an expression of continuity on some temporal order(s) of scale. Absolute continuity is characterized by dynamic activity in the system throughout the predictive span (the system exhibits “dense activity” of partially superimposed active unit populations as it’s state evolves). Effective absolute continuity is sustained when temporal gaps in activity are flanked by equivalent (activity) states prior to and immediately following gaps, i.e. state reinstating correlations sustained across apparent gaps.

Prediction and actual gating and direction
A simplistic partial model of a prediction-biased network exhibiting dynamics that are determined by actual input patterns

Anticipation is fuzzy. As input is absorbed the system’s state settles into favored unambiguous realizations that best match unfolding reality, and that continue to exert multiple-order cohesive and associative effects upon evolving anticipation. Another way to informally express this would be to say that the system is designed to decompose patterns into separable, spatially encoded, factors (sub patterns or operators) while at the same time it is driven to weave a contextually matched active expression of these related pattern-factors into its continuous dynamical trajectories. Because prediction is probabilistic it continuously anticipates a super-set of likely state trajectories, that dynamically affect it’s ‘impedance landscape’ and it therefore tends to ‘resonate’ into an input matching position from it’s state repertoire. ‘Lower frequency’ information encoded and enacted by anticipated or actual pattern factors must exert it’s influence on the trajectories of system state over longer spans of time (the supporting mechanism may not in itself be continuous, it could be stochastically recurrent, or somewhat blurred in time and recurring in a sequence).

In other words, No matter how much information regarding some environmental dynamic one collects there will always be a limit, an horizon, to predictive capacity that falls short relative to an ideal system. The missing  sub-sensitivity-threshold, sub-resolution, out-of bandwidth components determine the extent of predictive capacity shortfall. The smallest missing factors express the slowest to evolve, lowest frequency components, which determine the dynamic occurring just a little beyond the predictive horizon (and which can dramatically alter the state of the system by triggering non-linear effects). Even further from the horizon lie occurrences having zero limit effects on the perimeter, so their contribution to predictive potential, upon the perimeter span, decays to null. In this setting, an ideal agent’s system must be most receptive to probabilistically anticipated patterns but also infinitely agile to re-align with/’tune into’ forming surprise patterns (sensitivity to deviation from the anticipated). The tuning capacity of the system must rely acute sensitivity and quick, pervasive, reactivity to interaction with the world, mediated by sensory receptors on the phenomenal perimeter.

With respect to system tuning and accommodation of sensory information, applying the aforementioned influence of longer acting information (over time) implies that instantaneous system state expresses an aspect of computational temporal binding as on-going spatial (i.e. synchronous) configurations of related activity. i.e. momentary state is comprised of statistically relatively stable spatial configurations of activity (isotropic throughout a momentary interval), expressing relations and representing patterns of various duration, as these unfold from the past into the future. Such activity can be described as the projection of past–>future patterns around the present transition line, or as the realization of present-crossing patterns converging upon the present to drive ongoing transition. This temporally extended ‘smearing‘ effect, along with a notion of tonic core activity, is central to an hypothesis regarding the formation of ‘I’, which is explored in the next section.

Iconic pattern factors emulate real-time capture/play-out throughout a range of durations, from the shortest recordable momentary, to the longest behaviorally characterizing (partly overlapping) temporal span, all in continuous evolution. The continuous process applies to pattern factors in a continuum of temporal (and spatial) orders of scale. Lower orders make up the activity content of higher orders (i.e. nest in); Higher orders consist of modulations on lower order factors. These are essentially equivalent expressions of cross influences over several orders of scale. In the biological model, several distinct mechanisms act on a range of orders of scale to extend pattern representation capabilities.

Continuity is expressed by the activity patterns of populations of units either as incrementally changing ensembles (absolute continuity may not be mandated if incremental evolution is maintained across activity breaks, but it may possibly be conducive to phenomenal binding) or as switching of ensemble activity (not absolutely continuous).

High sensitivities, sensitivities to particular categories of change, broad tuning (i.e. low ‘Q’), overlapping receptive fields, time constants for natural decay, and a capacity to contain discontinuities (singularities) within patterns – all serve to promote continuity in system state evolution. Center/surround tuning – that seems to be ubiquitous in biological neural systems – may have evolved because of the advantage it confers to the ability to represent and maintain spatiotemporal segregation, contain edges and encode discontinuities. The push-pull arrangement facilitates the embedding of harmonic relations. On/off and opponent sensitivities may serve to mitigate loss of specificity due to the other mechanisms that maximize continuity, and perhaps most importantly to promote oscillations – maximize differentiated constrained modulated activity – in the system, in response to balanced on-off stimulation.

Highest order representational cohesion is mediated by contextual affinities associating several orders of complex representation. System level resources and capabilities such as intentional/attentional handles, i.e. something like ongoing re-generating activation of patterns in tandem with memory, may be relied upon to facilitate further extension of pattern replay capacity beyond the duration achievable with modular function, yet with decreased precision (see sub-section “the low frequency challenge” below).

A main tenet of the theory associates the manifestation of phenomenality with synchronous, isotropic ‘dense activity‘ over time. Consciousness exists in the limit of the ever shifting now. Momentary phenomenal manifestation (experiential contents of consciousness in a particular moment) is correlated with momentary activity in an ‘effectively absolutely continuous’ system. In a (proto-) phenomenal world it follows that spatial binding is effected by synchronous ‘dense activity’ and computational temporal binding manifests as phenomenal temporal binding.

It seems to be worth noting here some implications regarding conscious vs non-conscious states:

  • By the above, conscious states would be characterized by effectively-continuous, synchronous, isotropic configurations of temporally overlapping ‘dense activity’ in computational unit populations diffusely associating every-which way.
  • Non-conscious computational/pattern processing states, in contrast, would be characterized by non-continuity and temporal isolation/containment:
    1. Sparse, asynchronous, un-aligned, phased spread of activity or
    2. Minimally associated coherent activity (i,e, minimizing contextual interaction with extra duty cycle pattern processing).

These ideas are discussed further the most speculative section relating Computation and Phenomenality, which also includes some exploration of the philosophical perspective regarding temporal binding and the subjective experience of specious present.

Continuity may not always be maintained. In some moments a system may exhibit content loss, or untraceable fast forming (partial) context switches, and the corresponding phenomenal breaks. But through many moments anticipated states activated through hard-wired network connectivity and previously learnt Bayesian affinities are aligned with evolving state.

The main ideas presented in this work are meant to apply at orders of scale well above the physical classical limit. The extension of activity patterns such as suggested in this work to quantum phenomena orders of scale may, in some way, be plausible, with unfolding consequences in those orders of scale. Section three of this work ‘Computation and phenomenal content – Structure and materials‘ explores conceivable bridges to, and constraints upon, proto-phenomenal manifestation assuming order of scale segregation.  

Encoding and activation: Pattern embedding and activation modes

Encoding involves self organizing factorization, the results of which are subjected to an isomorphic (invertible, enactable) translation from the spatiotemporal domain into compressed-factor and cross-factor relata and relation representation in a (temporally invariant) spatial medium, which begets the term ‘Embedding’.

This term seems even more fitting when one considers that it does not suffice to encode inter-point associations enabling a playout of recorded associations in sequence. Dynamical attractor chasing, mediated through specific inter-point spatiotemporal affinities, does follow specific patterned event instances, but enacted representations must be complemented by a most important ‘bootstrapping’ function. This claim may seem obvious with regards to the encoding and activation of elementary pattern factors – for such elements that are encoded into a single modular unit where the isomorphic invertibility/enactability requirement entails the ability to ‘fire up’ the pattern-factor, but the claim also extends to the encoding and activation of composite patterns. Abstract, conceptual or semantic encoding and enactment of composite patterns can and should be efficient: At least in some modes of activation (e.g. anticipatory, intentional-semantic) multiple or all pattern factors contributing to a composite representation may be driven in tandem, all directly or indirectly feeding into, and being fed, in the correct pattern matching direction of association. I.e. multiple phases in respect of the duration of the pattern can be active concurrently, contributing to continuous ‘dense activity’.

Encoding may be attained via the formation of embedded pattern factor associations spanning a distinct spatial configuration, that when activated exhibits a tendency to isotropic activity throughout the configuration – i.e. as single point activity or synchronous (coherent) oscillation throughout a population, or ‘in-the-pattern’ wideband rectified noise (‘dense activity’). Such a configuration would be a temporally invariant isomorphic representation of that which would project a particular spatiotemporal stimulus pattern upon the sensory perimeter i.e. The information is embedded in the spatial ‘shape’ of potential isotropic population activity.
Once activated, the temporal characteristics of activation within the bounds of the spatial configuration are information poor. Morphing of the configuration of isotropic activity corresponds to (potential and actual) pattern evolution (change and composition).

Newly embedded patterns are likely to be represented by activity in a population of slightly differentiated module units. As a representation ‘consolidates’ some units become more differentiated i.e. more strongly linked with the pattern’s intrinsic and contextual cues, while other units become superfluous (e.g. if they fire later in a network exhibiting STDP ) and may take on other representations.

Due to the outlined principle of reciprocity in isomorphic transformation [pattern<==>isotropic activity], a multilevel cascade or a recursive application of isomorphic transforms (upon the respective isotropic activity product of encoding) may serve to increase the degree of [integration<==>(segregation + relation)] throughout the system (i.e. higher order representations combined through something like, metaphorically,  ‘blending’).

By way of implementation notes, encoding may be realized as an auto-sustaining mapping interference between a to-be-encoded dynamic signal pattern and a minimal set of encoding-driven orthogonal oscillations (i.e. locally spatially invariant, temporally repeating cycle), or finally – noise (progressing through a diffusing array of parallel and sequential cascades of sub-pattern processing). Embedded pattern factors are removed from the signal as encoding units fire reciprocally, so contributing to the interfering orthogonal reference. The ‘shape’ of the residual invariant active (isotropic) volume qualifies the embedded coding of a pattern.

Encoding must apply to and associate sub-patterns (and then factors) across all frequencies (or scale, resolution) in a source pattern’s recognizable and re-constitutable spectrum.

What is brought to the mix and how it is sorted…
The elucidation of interactions between signal patterns that originate in multiple sensory points is pragmatically facilitated by topological proximity of their respective signal paths, given that the system cannot exhibit all-to-all connectivity and is engineered to optimize continuity detection and population excitation preservation (e.g. as with small-world connectivity).

The low frequency challenge and spatial encoding

The fourier spectrum of slowly evolving patterns is comprised of low frequency components. These bottom-end component frequencies, requiring sensitivity close to the threshold of detectable difference between adjacent system states would be most challenging for a system to trace (and enact), because of cumulative errors and reduced signal to noise performance. Missing  sub-sensitivity-threshold, sub-resolution, out-of bandwidth potential sensory components determine the extent of predictive capacity, the limits of which are approached with exponentially diminishing returns on additional sensory input. The smallest missing factors express the slowest to evolve, lowest frequency components, which determine the dynamic occurring just beyond the predictive horizon. The lowest modelled/reconstitute-able frequency components limit the system’s predictive span, constraining it’s temporal “computational horizon”.

To best realize the predictive capacity supported by the system’s resolution and sensitivity it needs to be efficient and error-resilient. It also needs to optimize information conduction by supporting local processing, in proximity to the sensory perimeter, wherever such processing may be consistent with the agent’s global needs. Whether locally processed information is compressed and fed forward or perhaps utilized ‘in-situ’ by an ‘inverted system’ is central sub-text theme running through this work, that is explicitly explored in the section computation and phenomenality. Regardless, and claimed with relative confidence, the slowest to evolve (and most globally applicable) computational components require integration/differentiation over extended temporal and sensory spans.

In this context, it seems unlikely that the speed of active neural propagation has remained relatively slow because of some evolutionary barrier resistance. It seems more likely that slow synaptic dynamics were an essential characteristic and slow conduction was an advantageous evolutionary development, especially with larger brains. The conferred advantage endowed brains with the ability to literally trace bigger portions of the spatial trajectories of (spatiotemporally) patterned signal paths, associating signals presenting at different time points, while maximizing the distribution of processing and of the resources it consumes.

In a spiking network, the (relatively slow) propagation speed of neural conduction supports natural spatial tracing of evolving activity interaction patterns within the volume of the neural medium, while also keeping the action quanta units of operation distinct and spatially distributed. This localizes dynamical activity, reducing global processing load. It also reduces the agent’s (neural system) timing functions’ reliance on the precision of peri-synaptic dynamics (related to intrinsic composite time-response functions) in the presence of excitation-inhibition balance perturbations (e.g. implementations of predator-prey type dynamics), or on feedback loops involving physics of external or bodily motion, and motor action. It facilitates capture of complex patterns and extends the low end of the frequency spectrum that can be reliably encoded/reconstituted.

Activity quanta comprising sub-patterns at different spatial or temporal resolutions summate or combine to reconstitute/filter/match composite source signal patterns (some combinations are super-additive or otherwise modulated as they are directly enacted through non-linear filters or indirectly enacted via specialized function operators).

In the biological realm specialized mechanisms have evolved to extend the predictive horizon of agents’ systems by introducing embodied delay lines or controlled resistance to activity convergence towards attractors (tracing of even more slowly evolving patterns requires cues that do not exclusively rely on neural activity dynamics. These must be associated either with ecologically modulated slow rhythms that may be expressed through sensory coupled neural activity e.g. feedback from motor action, sensory input from cardiac rhythm, breathing actions and bowel movements, or the much slower circadian photoentrainment of photosensitive retinal ganglion cells, or with slow paced humoral or epigenetic modulators).

Dynamic Filters, the utility of phenomenal origin-point noise

Given externally driven perimeter stimuli, a filtering function can be substituted as an equivalent to a reconstruction or an enactment of a stimulus-matched anticipatory projection. The system dynamically adjusts it’s local “impedance” so as to effect the highest point “impedance” in anticipation of hard-wired or learnt state evolution; thereby facilitating efficient and consistent pattern matched “resonance”.

With sufficiently noisy sensory interface sources, filtering can be applied to simulate learnt patterns in the absence of external sensory input. Noise input can be enhanced through tonic activation with sensory feedback. Particular input can be further enhanced through phasic activation via feedback loops wherever appropriate efferent pathways and motor or filter capacities exist (e.g. outer hair cells of the inner ear).

Ambient, content-less consciousness could manifest as a flow of variously generated modulations over a noisy perimeter, a topic elaborated further in a following section on ‘Self’.

Side note: Scaled coherence, spectral scales and the precision required for maintaining relative phase relations

Spectral signal integrity can be maintained through factorization and reconstitution if phase relations (i.e. interference patterns, beat) between adjacent frequency components can be recorded and effected, respectively.

Overall signal integrity, i.e. coherence, can be maintained Coherence and precision relative to scalethrough preservation of relative phase relations with frequency band related precision. Relative phase between components of two different frequencies can be captured if the precision of the isomorphic relation (i.e. of the recorder/effector) is better than some fraction of the shorter wavelength (i.e. error is shorter then some fraction of the faster cycle).

Obviously phase relations between components are constrained to those frequency components that lie within the spectral range of interest, thus the required capture/enact precision is matched to the frequency band of interest.

However, with regards to what lies ‘internally’ to a sensory/phenomenal perimeter, it may be plausibly hypothesized that different frequency scales play at different ‘depths’ within the conscious agent’s machinery, interact with varying, mostly monotonically reducing, spans of the perimeter (i.e. locally – relative to scale), and therefore likely to be characterized by different precision requirements.
A unit, e.g. a cell, can only experience patterns that differentially affect points on it’s perimeter. This is true for patterns on any perceivable scale. A piece of membrane may only react to what impinges on it’s extent, an ion channel to the effects of the ionic flux across the protein units more or less spanning the width of the membrane etc. Conversely, a sensory surface may not posses resolving ability on the sub-cellular level; the filtering (and possibly transformation) of information both isolates and bridges bounded reactive, transformational or generative ‘information spaces’ (which in turn could correspond to (proto-) phenomenal manifestation compartments, see following sections).

The patterned dynamics of physical processes featuring on the perimeter do associate across many orders of scale: With phase relations maintained within band spans, and envelope modulation applied across bands that are further apart.

Since some ideas considered in the work involve correlations between loci of synchronous dense activity, it is pertinent to note that depending on the method and function chosen for practically computing ‘coherence’, rectified noise may exhibit a degree of ‘coherence’ with any other contemporaneous rectified signal. This inherent baseline degree of coherence must be compensated for, to differentiate stochastic activity modulated ‘in the patterns’ from ‘just noise’.

Author’s comment extending the preceding paragraph; Re-phrased applying poetic licence and personal bias – Coherence of ideally isotropic dense activity, pattern-less noise, could in some way be related to what raw, ambient, content less, consciousness feels like.

<-Prev                    Next ->

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s