Learning Algorithm Of Biological Networks

Artem Kirsanov
May 7, 2025
6 notes
6 Notes in this Video

Why Backpropagation Violates Core Biological Constraints

Backpropagation BiologicalPlausibility CreditAssignment NeuralDynamics
03:10

Artificial neural network researchers rely on backpropagation with gradient descent as the workhorse learning algorithm, but neurophysiologists point out that its requirements clash with how real neurons communicate and adapt in living brains.

NMDA Receptors as Molecular Coincidence Detectors

NMDAReceptors CoincidenceDetection HebbianLearning CalciumSignaling MolecularMechanisms
08:25

NMDA receptors function as voltage-sensitive calcium channels at excitatory synapses, implementing biological coincidence detection that underlies Hebbian learning through their unique dual-gating mechanism requiring both presynaptic glutamate release and postsynaptic depolarization.

Predictive Coding as Continuous Energy Minimization in Neural Circuits

PredictiveCoding EnergyBasedModels ErrorMinimization NeuralDynamics
11:40

Predictive coding theorists model the brain as an energy‑based system where neural activity relaxes toward states that minimize prediction error, providing a biologically plausible alternative to backpropagation for credit assignment.

In Vivo Imaging of Synaptic Plasticity During Learning

TwoPhotonImaging SynapticImaging MotorLearning FluorescentReporters SpineDynamics
16:45

Neuroscience researchers monitoring awake behaving mice learning lever-press tasks deployed dual-color fluorescent reporters to simultaneously track synaptic inputs (glutamate-sensitive “glu-sniffer” in green) and neuronal activity (calcium indicator R-CAMP in red) across consecutive days, revealing which activity patterns predict spine growth.

Compartment-Specific Plasticity Rules in Dendrites

DendriticCompartments LocalPlasticity NonHebbianLearning SynapticSegregation MotorCortex
20:10

Apical dendritic tuft compartments of layer 5 motor cortex pyramidal neurons exhibit fundamentally different plasticity rules than basal dendritic regions, with tuft plasticity driven by local dendritic activity independent of somatic spiking—contradicting classical Hebbian assumptions that require temporal correlation with postsynaptic action potentials.

Predictive Coding Implementation Through Dendritic Architecture

PredictiveCoding DendriticComputation ErrorSignals TopDownFeedback CorticalCircuitry
22:30

Cortical pyramidal neurons with their spatially segregated dendritic compartments appear optimized for predictive coding algorithms, where basal dendrites carry bottom-up sensory evidence while apical tufts receive top-down predictions from higher cortical areas, with dendritic nonlinearities computing prediction errors locally.