Brain theories {neural modeling} {brain theory} {computational neuroscience} model membrane currents, chemical changes, network oscillations, microcolumns, macrocolumns, and cell configurations to study learning and memory (Eric L. Schwartz).
viewpoint dependence
Real vision stores different viewpoints and matches similar schema.
single-neuron models
Neuron membranes have fast sodium-ion out-currents and later slower potassium-ion in-currents, as well as calcium ion, chloride ion, and other chemical flows, which affect action potential, adaptation, and shunting.
Dendrites and axons have structures and patterns.
Synapses have ion and chemical flows.
synapse plasticity
Synapse structure and physiology change over time with electrical and chemical flows. Feedback can alter weights in Hebbian learning.
More stable synapses learn and forget slower. Less stable synapses learn and forget faster. Systems use slow and fast plasticity combinations.
neural coding
Neurons have preferred stimuli. Neural coding can use instantaneous or average impulse frequency for rate coding. Impulses can code intensity, which can represent stimulus amplitude or stimulus probability.
Neuron signaling uses minimal number of impulses to convey information (Horace Barlow).
neural inhibition and excitation
Inhibition can subtract or divide. Excitation can add or multiply. Adding and subtracting accumulate same stimulus type, to pass or not pass threshold and determine whether to perform action. Multiplying and dividing represent stimulus interactions and feature pairing, to allow object detection or recognition.
neuron development
Neurons, axons, and dendrites migrate and grow. Migration and growth use hormone and growth-factor chemical gradients. For efficiency, wiring patterns are optimal in spacing and number {minimal wiring hypothesis}.
Sense physiology uses Bayesian inference, to reflect conditional rules.
neural networks
Neurons connect specifically to each other and use recurrence. Models use neuron pairs.
memory
Memory can be associative or content-addressable. Hippocampus models are for long-term memory. Prefrontal-cortex models are for working memory. Memory can use phase synchrony and wave resonance.
Electrical cables have resistance and capacitance, which determine oscillation time and dissipation length. Partial differential equations (William Thompson) are similar to wire heat-conductance equations (Fourier). Dendrites, cell bodies, and axons are like cables and have capacitances and resistances in parallel and series {cable theory} (Wilfrid Rall). Fibers have resistance because cytoplasm and membrane are not good conductors. Fibers have capacitance because membrane phospholipid bilayer does not conduct but has charge polarization.
Functions {Gabor function}, derived from neuron response frequencies, can represent neuron location, width, length, orientation, and frequency as wavelet. Gabor functions represent neuron types. Neural nets or systems are Gabor-function configurations.
Units representing neurons can have binary outputs off or on, use input thresholds, and be in networks. Neural networks {Hopfield net} (John Hopfield) can use recurrence to iteratively determine final values.
convergence
Outcomes are locations and have values. Hopfield nets converge on local minima among set.
training
Training with images can determine locations that represent standard features or objects.
recognition
If local minimum matches location representing feature, Hopfield net recognizes feature in images. In particular, input that is only feature index or cue can lead to the feature, so Hopfield nets can act like memory system {content-addressable memory, Hopfield net}.
Statistical-mechanics models {Ising model} can use pairs of +1 or -1 spins (Ernst Ising) [1925]. Pair spins can have same or different alignment. Pair energy is product of spins: +1 * +1 = 1 = -1 * -1, or +1 * -1 = -1 = -1 * +1.
Network can show spin interactions. Nodes are particles with spins, and connection edges are interaction energies. Regions, and whole systems, have total energy and are system states.
Neural nets have binary units, +1 and -1, as nodes and have connections among neurons. Total energy is system state.
3-Computer Science-Systems-Computer Vision
Outline of Knowledge Database Home Page
Description of Outline of Knowledge Database
Date Modified: 2022.0225