Processor emulation is a virtualization technology that allows software compiled for one processor & operating system to run on a system with a different processor & operating system, without any source code or binary changes.
KorrTecx is an emulation of a biological processor, the AI runs on the simulated biological processor, not the Von Neumann based architecture.
Hardware - PC Cluster
Hardware comprises of twelve four core PC nodes in a standard cluster. The Nodes are old-skool 4 core Q6600's OC @ 3Ghz with SSD.
Compute Nodes
The processing core is designed around a modular 'compute node' architecture governed by a custom MPI. Each node runs on a single core and can be configured for either a segmented large model or many small models with differing parameters for testing.
Sensory Hardware & Modules
Vision - Bot currently has two axis of movement & three HD cameras, two for peripheral stereo with a 30 degree overlap & one for the high resolution fovea. [Video]
Auditory - Stereo microphones provide audio input and vocalisation is provided a speaker. Accelerometers register physical movement and correlate feedback.
Tactile - The tactile sensory streams used in my videos are extrapolated from actual data recorded from an analogue source.
Bespoke 4D CAD & simulation engine, a mesh of 3D guides defines the spatial characteristics of the neural structures and laminate layers. An embedded octree mesh enables volumetric bio-electrochemical functionality.
Algorithms simulate transmitter & modulator gate uptake, dissipation, accumulation, neuro/ synaptogenesis, infinite neuron types & arbour descriptors, maturation, growth, sleep & many other bioprocesses.
Example background algorithms for the biological simulation.
Neuron & Synapse Culling - Remove redundant.
Spike Potentials - Spatiotemporal simulation spikes.
Voxel Transmitters - Uptake, dissipation rates and types.
Neuron & Synapse Maturation - cell growth & death.
EM field Generation - Not used at the moment.
Transmitters - Dissipation in Interstitial fluid.
Maturation Timeline - Development & growth of the model.
Sleep Cycles - Scripts for controlling the mechanisms of sleep.
Example evolutionary algorithms.
Neurogenesis - Adds neurons relative to local requirements.
Synaptogenesis - Adds synapse relative to local requirements.
Dendrite Guidance - Extending along transmitter profiles.
Axon Guidance - Growth cones follow transmitter profiles.
Self Organisation - Cell organisation (cortical columns).
Self Optimization - network & cluster optimization.
Dendritic Processing
Gate Transmitter Uptake & Dissipation
The models morphology is roughly based on the human template. The scale and location of the deep brain structures align with academia's generic template. Initial long range tract guides between the structures follow empirical data from a human connectome.
Connectomics are defined by evolutionary algorithms that evolve & adapt the model over time. Long-range inter-structure (white matter) axons, the medium/ short-range (grey matter) tracts within the semantic laminate layers & localised structures like dendritic arbours and hubs are all simulated.
Structures - Neocortex (gyrification), lobes, hippocampus, thalamus, cerebellum, stem (pons) & wider nervous system (homunculus map)
Cortices - Motor, pre-motor, somatosensory, pre-frontal, frontal, visual (1,2,3,4), audio
Connectomics - Mini maps, hubs, clusters, tracts, semantic maps, cortical columns
Substrate - Laminates (six layer), neurons, dendrites, synapse, axons
Bio-chemical - Myelin, transmitters, neuro modulators
Bio-electrical - Spatiotemporal spikes, EM fields
Structures & functionality
The overall morphology is based roughly on the human template. The scale and location of the deep brain structures align with academia's generic template. Initial long range tract guides between the structures follow empirical data from a human connectome.
Manifold - shape optimal for thalamocortical rhythms.
Lobes - The lobe boundaries key role is halting surface propagation between diverse functional areas.
Gyrification - gyri/ sulci (GS) are not just the result of a limited volume & expanding cortical sheet. There is a direct relation between GS and the other deep structures.
The cortical folding pattern (Gyri/ Sulci) not only allows for a greater semantic surface area but also a broader frequency/ phase response per unit area.
Substrate - Six layers
Laminate Layers - Lateral tuning
Semantic Maps - In an evolved connectome model visual, auditory & tactile senses all share generalised/ semantic constructs, eg. stereo disparity... we can 'see' with touch, etc. Its one external spatiotemporal sense (thalamic modulation), the grounding differentiates the modalities.
Sparse networks of distributed neuron clusters (NC) create generalised semantic maps.
The versatility of human cognition is derived from the re-combination of these NC's. A single NC can contribute properties to millions of diverse memory engrams.
Cortical Columns - Mini columns are innate, macro cortical columns are not.
They are a product of tuning/ pruning of lateral (laminar) dendritic branches during learning.
Inhibition - localised lateral inhibition within the laminar structures, sharpens semantic map response by dampening the local, competing areas, allowing the strongest (gradient ascent) signal to dominate. This regulates global signal sparsity/ homeostasis.
Cortices - Apparent distinct cortical regions (excluding sensory input) with specific functionality (FFA, Broca) are NOT directly processing related information. Their location is optimal for tuning, to extract the relevant information from the globally distributed processing.
Frontal Lobe - a special case/ area, no direct association to any sensory areas, stimulated by long-range afferent axons that originate in the hubs. Purely processes internal activity generated by the rest of the connectome & influences/ conducts the global narrative.
Mini-maps (MM) - lateral self-organised freq response across the cortical maps (neo) tends to leave areas of low connectivity, MM are localised columns which form in these areas. Connected by long lateral tracts they provide a sparse blended sum of local activation.
Hubs - areas where long range tracts terminate, used for fast signal transmission and balancing global BTR homeostasis.
Clusters - smaller than hubs or association areas/ maps, form where multiple distributed networks share/ have a similar frequency space.
An overloaded neuron basically calls for help, and neurogenesis sends reinforcements to divide the workload.