Hopfield networks

E46142

associative memory model content-addressable memory system recurrent artificial neural network

Hopfield networks are recurrent artificial neural networks that serve as content-addressable memory systems, storing patterns as stable states and retrieving them through dynamics that minimize an energy function.

Aliases (3)

Statements (50)

Predicate	Object
instanceOf	associative memory model → content-addressable memory system → recurrent artificial neural network →
belongsToField	computational neuroscience → machine learning → neural networks → statistical physics →
convergesTo	local energy minima →
dynamicsMinimize	energy function →
hasActivationFunction	sign function → threshold function →
hasApproximateCapacity	0.138N for random uncorrelated patterns →
hasCapacityProperty	storage capacity proportional to number of neurons →
hasConnectionType	no self-connections → symmetric weights →
hasEnergyFunction	Lyapunov function →
hasLearningRule	Hebbian learning → outer-product rule →
hasLimitation	limited storage capacity → sensitivity to correlated patterns → spurious attractors →
hasMathematicalRepresentation	binary quadratic form energy →
hasNodeType	Ising spin → binary neuron →
hasProperty	guaranteed convergence under symmetric weights and asynchronous updates →
hasStateSpace	binary vectors →
hasTopology	fully connected network →
hasUpdateDynamics	deterministic dynamics →
hasUpdateRule	asynchronous update → synchronous update →
hasVariant	continuous Hopfield network → modern Hopfield network → stochastic Hopfield network →
introducedBy	John Hopfield →
introducedInYear	1982 →
isRelatedTo	Boltzmann machine → Ising Hopfield model → Ising model → spin glass theory →
isUsedFor	associative memory tasks → combinatorial optimization → constraint satisfaction → optimization →
namedAfter	John Hopfield →
stableStatesRepresent	stored patterns →
supports	associative recall → content-addressable memory → error correction → pattern completion → robust retrieval from noisy inputs →

Referenced by (3)

Subject (surface form when different)	Predicate
Hopfield network ("continuous Hopfield network") → Hopfield network ("modern Hopfield network") →	hasVariant
Boltzmann machines →	relatedTo