The Computational Brain

Knowledge Graph: The Computational Brain (Patricia S. Churchland & Terrence J. Sejnowski, 1992)

Editorial spotlight: ↑ the co-evolutionary constraint satisfaction loop

Concepts

• C&S co-evolutionary constraint satisfaction (importance 5): Understanding requires iterating between four levels simultaneously — computational theory, algorithm, implementation, and evolutionary constraints — with no privileged starting point.. Source: (from training memory of book).
• Marr's three levels of analysis (importance 4): Computational theory (what/why), algorithm/representation (how), hardware implementation (physical substrate). Foundational framework but insufficient without evolutionary history.. Source: (from training memory of book).
• nervous energy budget constraints (importance 4): Brain represents ~2% body mass but ~20% energy budget in humans; this drives architectural choices toward efficiency.. Source: (from training memory of book).
• C&S sparse distributed coding (importance 4): Information encoded in patterns of activity across populations where only small fraction active at once; energy-efficient and fault-tolerant.. Source: (from training memory of book).
• C&S neural state space (importance 4): Network dynamics viewed as trajectory through high-dimensional activation space; attractors correspond to stable representations.. Source: (from training memory of book).
• C&S vision as inverse optics (importance 4): Visual system solves ill-posed inverse problem of inferring 3D world from 2D retinal images using constraints and assumptions.. Source: (from training memory of book).
• C&S what vs. where streams (importance 4): Ventral stream (what/object recognition) vs. dorsal stream (where/spatial relations); double dissociation in lesion studies.. Source: (from training memory of book).
• C&S synaptic learning rules (importance 4): Hebbian, anti-Hebbian, and covariance rules modify connection strengths based on correlated activity patterns.. Source: (from training memory of book).
• C&S cortical self-organization (importance 4): Structured maps emerge from unsupervised learning driven by input statistics and lateral interactions.. Source: (from training memory of book).
• C&S hippocampal memory model (importance 4): Hippocampus as fast-learning temporary store enabling consolidation to neocortex; complementary learning systems theory.. Source: (from training memory of book).
• C&S efficient coding hypothesis (importance 4): Sensory systems evolved to maximize information transmitted about natural environment subject to metabolic constraints.. Source: (from training memory of book).
• C&S neural network methodology (importance 4): Simplified models reveal computational principles; test algorithmic ideas; guide experimental predictions.. Source: (from training memory of book).
• computational level (C&S sense) (importance 3): Abstract specification of what function is computed and why it's appropriate to the organism's ecological niche.. Source: (from training memory of book).
• algorithmic level (C&S sense) (importance 3): How the computation is carried out — the representations and processes that transform inputs to outputs.. Source: (from training memory of book).
• implementation level (C&S sense) (importance 3): Physical realization in neurons, synapses, neural circuits — the wetware that instantiates the algorithm.. Source: (from training memory of book).
• cranial volume constraint (importance 3): Skull size limits total neuron count and wire length; explains compression strategies and local computation emphasis.. Source: (from training memory of book).
• population vector coding (importance 3): Vector direction encoded by weighted sum of broadly-tuned neuron responses; demonstrated in motor cortex by Georgopoulos.. Source: (from training memory of book).
• coarse coding theorem (importance 3): Many broadly-tuned units can achieve finer discrimination than narrowly-tuned units; resolution increases with population size.. Source: (from training memory of book).
• attractor network models (importance 3): Networks settle into stable states (point attractors) or oscillations (limit cycles) that represent memories or computations.. Source: (from training memory of book).
• climbing fiber teaching signal (importance 3): Error signal from inferior olive that triggers LTD at parallel fiber synapses; biological supervised learning mechanism.. Source: (from training memory of book).
• ill-posed inverse problem (importance 3): Underconstrained problem with infinite solutions; brain adds regularization constraints to achieve unique stable solution.. Source: (from training memory of book).
• regularization theory (Poggio/Torre) (importance 3): Adding smoothness constraints to ill-posed problems; explains why brain prefers simplest interpretation consistent with data.. Source: (from training memory of book).
• C&S stereopsis computation (importance 3): Depth from binocular disparity requires solving correspondence problem — which left-eye feature matches which right-eye feature.. Source: (from training memory of book).
• V1 orientation selectivity (importance 3): Simple cells respond maximally to oriented edges; emergent from LGN input arrangement and cortical circuitry.. Source: (from training memory of book).
• ventral 'what' pathway (importance 3): V1 → V4 → IT cortex; progressively invariant object representations; damage causes agnosia.. Source: (from training memory of book).
• dorsal 'where' pathway (importance 3): V1 → MT → parietal cortex; spatial location and motion; damage causes spatial neglect and optic ataxia.. Source: (from training memory of book).
• C&S motion detection models (importance 3): Reichardt detector and variants; spatiotemporal correlation of local luminance changes extracts velocity.. Source: (from training memory of book).
• Hebbian learning (importance 3): Fire together, wire together; strengthens synapses between coactive neurons; implements correlation-based plasticity.. Source: (from training memory of book).
• C&S dimensionality reduction (importance 3): Sensory systems compress high-dimensional input to lower-dimensional representations preserving task-relevant structure.. Source: (from training memory of book).
• ocular dominance column formation (importance 3): Alternating left-eye/right-eye stripes in V1 emerge from Hebbian competition for cortical territory during development.. Source: (from training memory of book).
• C&S retinal preprocessing (importance 3): Retina not just transducer but performs edge detection, motion detection, and adaptation before sending to cortex.. Source: (from training memory of book).
• CA3 autoassociative network (importance 3): Recurrent collaterals in CA3 implement content-addressable memory; pattern completion from partial cues.. Source: (from training memory of book).
• long-term potentiation (LTP) (importance 3): Activity-dependent strengthening of synapses; NMDA receptor mechanism implements Hebbian coincidence detection.. Source: (from training memory of book).
• dopamine reward signal (importance 3): VTA/SNc dopamine neurons encode reward prediction error; gates synaptic plasticity in striatum.. Source: (from training memory of book).
• C&S neural time scale hierarchy (importance 3): Different processes operate at different time scales: spikes (ms), synaptic integration (10-100ms), learning (minutes-years).. Source: (from training memory of book).
• C&S dendritic computation (importance 3): Dendrites are active computational units with voltage-gated channels; implement nonlinear operations beyond point neuron.. Source: (from training memory of book).
• C&S neural oscillations (importance 3): Rhythmic population activity (gamma, theta, alpha bands); may coordinate communication between brain regions.. Source: (from training memory of book).
• C&S neuromodulatory systems (importance 3): Diffuse projection systems (ACh, NE, 5-HT, DA) adjust neural gain and plasticity globally; regulate brain states.. Source: (from training memory of book).
• redundancy reduction (importance 3): Neural codes remove statistical redundancy in sensory input to enable efficient representation; decorrelates signals.. Source: (from training memory of book).
• sparse representation principle (importance 3): Represent inputs with few active units from large overcomplete basis; efficient and robust to noise.. Source: (from training memory of book).
• C&S predictive coding (importance 3): Brain predicts sensory input and encodes prediction errors; efficient representation and hierarchical inference.. Source: (from training memory of book).
• Bayesian inference hypothesis (importance 3): Brain performs probabilistic inference combining sensory evidence with prior expectations; optimal under uncertainty.. Source: (from training memory of book).
• C&S computational motor control (importance 3): Movement emerges from hierarchical control: high-level goals → trajectories → muscle synergies → motor neurons.. Source: (from training memory of book).
• cerebellar LTD (importance 2): Long-term depression at parallel fiber synapses when climbing fiber fires; Ito's discovery of bidirectional plasticity.. Source: (from training memory of book).
• stereo correspondence problem (importance 2): Ambiguity in matching features across two eyes; resolved by uniqueness and continuity constraints.. Source: (from training memory of book).
• receptive field structure (importance 2): Region of sensory space where stimulation affects neuron; defines what feature the neuron detects.. Source: (from training memory of book).
• Gabor filter model of V1 (importance 2): Simple cells as localized oriented Gabor wavelets; good match to physiological data and mathematically principled.. Source: (from training memory of book).
• spatial frequency channels (importance 2): Visual system decomposes image into multiple scale bands; parallel processing of coarse and fine structure.. Source: (from training memory of book).
• aperture problem (importance 2): Local motion measurement ambiguous for oriented edges; requires integration across space to recover true velocity.. Source: (from training memory of book).
• competitive learning networks (importance 2): Winner-take-all dynamics with Hebbian strengthening; leads to emergence of feature detectors and clustering.. Source: (from training memory of book).
• PCA network models (importance 2): Hebbian networks can implement principal component analysis; extract directions of maximum variance in input.. Source: (from training memory of book).
• orientation preference map (importance 2): Smooth variation of preferred orientation across V1 with pinwheel singularities; emerges from activity-dependent wiring.. Source: (from training memory of book).
• Mexican hat connectivity (importance 2): Short-range excitation plus long-range inhibition; generates winner-take-all dynamics and sharpens tuning curves.. Source: (from training memory of book).
• lateral inhibition contrast enhancement (importance 2): Inhibitory surrounds sharpen edge responses and implement gain control; computational role in early vision.. Source: (from training memory of book).
• ON-center/OFF-center cells (importance 2): Complementary populations respond to light increments vs. decrements; retinal ganglion cell types preserving full luminance range.. Source: (from training memory of book).
• center-surround antagonism (importance 2): Receptive field with excitatory center and inhibitory surround or vice versa; implements edge detection and adaptation.. Source: (from training memory of book).
• corticothalamic feedback (importance 2): Massive top-down projections from V1 to LGN; modulates gain and implements attentional gating.. Source: (from training memory of book).
• color-opponent coding (importance 2): Red-green and blue-yellow opponent channels in retina and LGN; efficient coding of natural color statistics.. Source: (from training memory of book).
• direct/indirect pathway model (importance 2): Opposing striatal pathways facilitate vs. suppress actions; imbalance in Parkinson's disease.. Source: (from training memory of book).
• cable theory (importance 2): Mathematical model of signal propagation in dendrites; passive spread with exponential attenuation.. Source: (from training memory of book).
• active dendritic conductances (importance 2): Voltage-gated channels in dendrites enable backpropagating action potentials and local spike generation.. Source: (from training memory of book).
• action potential generation (importance 2): All-or-none regenerative spike via Na+ channel activation and K+ repolarization; digital signaling over long distances.. Source: (from training memory of book).
• synaptic integration (importance 2): Dendrites sum excitatory and inhibitory inputs; if depolarization reaches threshold at axon hillock, spike fires.. Source: (from training memory of book).
• shunting inhibition (importance 2): Inhibitory conductance reduces membrane resistance; divisive rather than subtractive effect on excitation.. Source: (from training memory of book).
• gamma oscillations (30-80 Hz) (importance 2): Fast oscillations in local circuits via interneuron networks; linked to attention and feature binding.. Source: (from training memory of book).
• hippocampal theta (4-8 Hz) (importance 2): Dominant rhythm during exploration and REM sleep; paces learning and memory consolidation.. Source: (from training memory of book).
• binding problem (importance 2): How brain integrates distributed features into unified percepts; temporal synchrony hypothesis.. Source: (from training memory of book).
• temporal synchrony binding (importance 2): Features of same object represented by synchronously firing neurons; oscillations provide temporal windows.. Source: (from training memory of book).
• natural scene statistics (importance 2): Statistical regularities in natural images (1/f power spectrum, sparse structure); shape sensory coding strategies.. Source: (from training memory of book).
• internal forward models (importance 2): Motor system maintains predictive model of sensory consequences of actions; enables feedforward control.. Source: (from training memory of book).
• sensorimotor coordinate transforms (importance 2): Converting between reference frames (eye, head, body, world); posterior parietal cortex implements basis function networks.. Source: (from training memory of book).
• basis function networks (importance 2): Distributed encoding using units with overlapping tuning curves; implement smooth nonlinear transformations.. Source: (from training memory of book).
• gain field modulation (importance 2): Multiplicative modulation of sensory response by postural signals; implements coordinate transformations.. Source: (from training memory of book).
• TD learning (Sutton) (importance 2): Reinforcement learning using prediction errors; dopamine resembles TD error signal.. Source: (from training memory of book).
• neural scaling relationships (importance 2): Brain structure scales predictably with body/brain size across species; allometric relationships reveal design constraints.. Source: (from training memory of book).
• retinal temporal filtering (importance 1): Different ganglion cell types extract sustained vs. transient features; parallel temporal channels to LGN.. Source: (from training memory of book).
• membrane time constant (importance 1): RC time constant of neuronal membrane (~10-50ms); determines temporal integration window.. Source: (from training memory of book).

Claims

• C&S levels must interact bidirectionally (importance 5): Understanding requires simultaneous constraint from theory, algorithm, implementation, and evolution — no level is privileged.. Source: (from training memory of book).
• C&S fourth level: evolutionary constraints (importance 4): Brain design reflects evolutionary history and selection pressures; understanding requires asking what ecological problems shaped the solution.. Source: (from training memory of book).
• wire length minimization principle (importance 4): Neural architectures minimize total axonal length while maintaining necessary connectivity; drives hierarchical and columnar organization.. Source: (from training memory of book).
• biological plausibility constraint (importance 3): Models must respect known neural constraints (time constants, connectivity, learning rules) to be explanatory.. Source: (from training memory of book).

Empirical results

• C&S cerebellar learning model (importance 4): Marr-Albus-Ito theory of motor learning via adjustable parallel fiber-Purkinje cell synapses; example of algorithm-implementation bridge.. Source: (from training memory of book).

Entities

• Hopfield network (1982) (importance 3): Energy-minimizing network that demonstrates content-addressable memory via symmetric weights and convergence to stable states.. Source: (from training memory of book).
• backpropagation algorithm (importance 3): Credit assignment through multi-layer networks via gradient descent; biologically implausible but computationally effective.. Source: (from training memory of book).
• NETtalk (Sejnowski & Rosenberg) (importance 3): Network learning text-to-speech mapping; demonstrated emergent hierarchical representations without hand-coded linguistic rules.. Source: (from training memory of book).
• vestibulo-ocular reflex adaptation (importance 3): Cerebellar calibration of eye movements to compensate head motion; classic example of motor learning requiring precise timing.. Source: (from training memory of book).
• Hubel & Wiesel hierarchy (importance 3): Simple cells → complex cells → hypercomplex cells; foundational model of feature detection hierarchy but now seen as oversimplified.. Source: (from training memory of book).
• basal ganglia action selection (importance 3): Subcortical loops implementing winner-take-all for motor program selection; dopamine modulates learning.. Source: (from training memory of book).
• Boltzmann machine (importance 2): Stochastic version of Hopfield net using simulated annealing; can learn hidden structure but computationally expensive.. Source: (from training memory of book).
• Purkinje cell computation (importance 2): Receives ~100k parallel fiber inputs and climbing fiber teaching signal; implements perceptron-like adjustable threshold.. Source: (from training memory of book).
• Marr-Poggio stereo algorithm (importance 2): Multi-scale cooperative algorithm using edge detection and disparity-tuned units; influential but biologically oversimplified.. Source: (from training memory of book).
• inferotemporal cortex (importance 2): High-level object representations; neurons respond to complex shapes/faces with position/size invariance.. Source: (from training memory of book).
• area MT (V5) (importance 2): Middle temporal area specialized for motion detection; direction-selective cells and speed tuning.. Source: (from training memory of book).
• BCM rule (Bienenstock-Cooper-Munro) (importance 2): Sliding threshold determines whether potentiation or depression occurs; models cortical plasticity during development.. Source: (from training memory of book).
• Kohonen self-organizing map (importance 2): Competitive learning with neighborhood function; creates topology-preserving maps like cortical sensory maps.. Source: (from training memory of book).
• lateral geniculate nucleus (importance 2): Thalamic relay preserving retinotopy; six layers with different cell types; more than passive relay due to cortical feedback.. Source: (from training memory of book).
• dentate gyrus pattern separation (importance 2): Sparse coding in dentate separates similar inputs; prevents interference in CA3 memory storage.. Source: (from training memory of book).
• NMDA receptor (importance 2): Voltage-dependent glutamate receptor requiring presynaptic release and postsynaptic depolarization; molecular Hebbian coincidence detector.. Source: (from training memory of book).
• hippocampal place cells (importance 2): CA1 neurons with spatially-selective firing; cognitive map representation discovered by O'Keefe.. Source: (from training memory of book).
• striatum (importance 2): Input stage of basal ganglia receiving cortical projections; medium spiny neurons integrate context for action selection.. Source: (from training memory of book).
• Hodgkin-Huxley model (importance 2): Quantitative model of action potential generation via Na+ and K+ conductances; foundational biophysical neuron model.. Source: (from training memory of book).
• acetylcholine system (importance 2): Basal forebrain projections enhance sensory processing and enable plasticity; depleted in Alzheimer's disease.. Source: (from training memory of book).
• primary motor cortex (importance 2): M1 contains somatotopic map; population vectors encode movement direction; outputs to spinal cord.. Source: (from training memory of book).
• posterior parietal cortex (importance 2): Multisensory integration and sensorimotor transformations; damage causes spatial neglect and reaching deficits.. Source: (from training memory of book).
• CA1 comparator (importance 1): Receives CA3 output and direct entorhinal input; may detect novelty by comparing predicted vs. actual patterns.. Source: (from training memory of book).
• entorhinal grid cells (importance 1): Hexagonal firing pattern tiles environment; may provide metric for place cell computation (discovered post-1992).. Source: (from training memory of book).
• norepinephrine (locus coeruleus) (importance 1): LC projects widely; modulates arousal, attention, and stress response; increases signal-to-noise ratio.. Source: (from training memory of book).
• serotonin (raphe nuclei) (importance 1): Raphe projections throughout brain; regulates mood, sleep, and impulse control.. Source: (from training memory of book).
• premotor cortex (importance 1): Motor planning and sensorimotor transformations; visual guidance of reaching.. Source: (from training memory of book).

Relations

• C&S co-evolutionary constraint satisfaction builds-on Marr's three levels of analysis
• C&S co-evolutionary constraint satisfaction requires C&S fourth level: evolutionary constraints
• Marr's three levels of analysis generalizes computational level (C&S sense)
• Marr's three levels of analysis generalizes algorithmic level (C&S sense)
• Marr's three levels of analysis generalizes implementation level (C&S sense)
• C&S fourth level: evolutionary constraints motivates nervous energy budget constraints
• C&S fourth level: evolutionary constraints motivates cranial volume constraint
• nervous energy budget constraints motivates wire length minimization principle
• nervous energy budget constraints motivates C&S sparse distributed coding
• cranial volume constraint motivates wire length minimization principle
• wire length minimization principle supports C&S cortical self-organization
• C&S sparse distributed coding generalizes population vector coding
• C&S sparse distributed coding exemplifies C&S efficient coding hypothesis
• population vector coding enables coarse coding theorem
• C&S neural state space generalizes attractor network models
• attractor network models exemplifies Hopfield network (1982)
• attractor network models exemplifies Boltzmann machine
• Hopfield network (1982) motivates CA3 autoassociative network
• backpropagation algorithm enables NETtalk (Sejnowski & Rosenberg)
• NETtalk (Sejnowski & Rosenberg) exemplifies C&S neural network methodology
• C&S cerebellar learning model requires Purkinje cell computation
• C&S cerebellar learning model requires climbing fiber teaching signal
• C&S cerebellar learning model evidences vestibulo-ocular reflex adaptation
• climbing fiber teaching signal enables cerebellar LTD
• cerebellar LTD exemplifies C&S synaptic learning rules
• C&S vision as inverse optics exemplifies ill-posed inverse problem
• ill-posed inverse problem requires regularization theory (Poggio/Torre)
• C&S vision as inverse optics exemplifies C&S stereopsis computation
• C&S stereopsis computation requires stereo correspondence problem
• stereo correspondence problem motivates Marr-Poggio stereo algorithm
• V1 orientation selectivity evidences Hubel & Wiesel hierarchy
• V1 orientation selectivity supports Gabor filter model of V1
• Hubel & Wiesel hierarchy exemplifies receptive field structure
• Gabor filter model of V1 supports spatial frequency channels
• C&S what vs. where streams generalizes ventral 'what' pathway
• C&S what vs. where streams generalizes dorsal 'where' pathway
• ventral 'what' pathway requires inferotemporal cortex
• dorsal 'where' pathway requires area MT (V5)
• area MT (V5) exemplifies C&S motion detection models
• C&S motion detection models requires aperture problem
• C&S synaptic learning rules generalizes Hebbian learning
• Hebbian learning generalizes BCM rule (Bienenstock-Cooper-Munro)
• C&S synaptic learning rules enables competitive learning networks
• competitive learning networks enables C&S dimensionality reduction
• C&S dimensionality reduction exemplifies PCA network models
• C&S cortical self-organization exemplifies Kohonen self-organizing map
• C&S cortical self-organization evidences ocular dominance column formation
• C&S cortical self-organization evidences orientation preference map
• competitive learning networks supports ocular dominance column formation
• Mexican hat connectivity enables lateral inhibition contrast enhancement
• Mexican hat connectivity enables competitive learning networks
• C&S retinal preprocessing requires ON-center/OFF-center cells
• C&S retinal preprocessing requires center-surround antagonism
• ON-center/OFF-center cells exemplifies center-surround antagonism
• center-surround antagonism enables lateral inhibition contrast enhancement
• C&S retinal preprocessing requires retinal temporal filtering
• retinal temporal filtering precedes lateral geniculate nucleus
• lateral geniculate nucleus requires corticothalamic feedback
• lateral geniculate nucleus exemplifies color-opponent coding
• C&S hippocampal memory model requires CA3 autoassociative network
• C&S hippocampal memory model requires dentate gyrus pattern separation
• CA3 autoassociative network requires long-term potentiation (LTP)
• dentate gyrus pattern separation precedes CA3 autoassociative network
• CA3 autoassociative network precedes CA1 comparator
• long-term potentiation (LTP) requires NMDA receptor
• NMDA receptor exemplifies Hebbian learning
• CA1 comparator exemplifies hippocampal place cells
• entorhinal grid cells enables hippocampal place cells
• basal ganglia action selection requires dopamine reward signal
• basal ganglia action selection requires striatum
• dopamine reward signal supports TD learning (Sutton)
• striatum generalizes direct/indirect pathway model
• C&S neural time scale hierarchy exemplifies membrane time constant
• C&S dendritic computation builds-on cable theory
• C&S dendritic computation requires active dendritic conductances
• active dendritic conductances builds-on Hodgkin-Huxley model
• Hodgkin-Huxley model evidences action potential generation
• action potential generation requires synaptic integration
• synaptic integration requires shunting inhibition
• C&S neural oscillations generalizes gamma oscillations (30-80 Hz)
• C&S neural oscillations generalizes hippocampal theta (4-8 Hz)
• binding problem motivates temporal synchrony binding
• temporal synchrony binding requires gamma oscillations (30-80 Hz)
• C&S neuromodulatory systems generalizes acetylcholine system
• C&S neuromodulatory systems generalizes norepinephrine (locus coeruleus)
• C&S neuromodulatory systems generalizes serotonin (raphe nuclei)
• C&S neuromodulatory systems generalizes dopamine reward signal
• C&S efficient coding hypothesis requires redundancy reduction
• C&S efficient coding hypothesis supports sparse representation principle
• redundancy reduction requires natural scene statistics
• natural scene statistics motivates sparse representation principle
• C&S predictive coding supports Bayesian inference hypothesis
• Bayesian inference hypothesis requires internal forward models
• internal forward models enables C&S computational motor control
• C&S computational motor control requires primary motor cortex
• C&S computational motor control requires C&S cerebellar learning model
• primary motor cortex exemplifies population vector coding
• primary motor cortex precedes premotor cortex
• sensorimotor coordinate transforms requires basis function networks
• basis function networks exemplifies posterior parietal cortex
• posterior parietal cortex exemplifies gain field modulation
• gain field modulation enables sensorimotor coordinate transforms
• C&S levels must interact bidirectionally supports C&S co-evolutionary constraint satisfaction
• C&S neural network methodology requires biological plausibility constraint
• biological plausibility constraint requires implementation level (C&S sense)
• neural scaling relationships supports nervous energy budget constraints
• neural scaling relationships supports cranial volume constraint
• C&S efficient coding hypothesis motivates C&S retinal preprocessing
• C&S sparse distributed coding exemplifies dentate gyrus pattern separation
• TD learning (Sutton) supports dopamine reward signal
• C&S neural state space motivates C&S hippocampal memory model
• competitive learning networks enables C&S cortical self-organization
• C&S co-evolutionary constraint satisfaction requires C&S levels must interact bidirectionally
• C&S vision as inverse optics exemplifies computational level (C&S sense)
• C&S cerebellar learning model exemplifies algorithmic level (C&S sense)
• Purkinje cell computation exemplifies implementation level (C&S sense)
• C&S what vs. where streams evidences C&S fourth level: evolutionary constraints
• C&S predictive coding motivates corticothalamic feedback
• C&S efficient coding hypothesis supports C&S co-evolutionary constraint satisfaction
• Hebbian learning enables C&S cortical self-organization
• C&S neuromodulatory systems enables C&S synaptic learning rules

© 2026 LOOMUS. Curation, importance scoring, edge typing, and visualization. Source paper © respective authors, quoted under fair use.