Oja rule
E899010
Oja rule is a normalized form of Hebbian learning used in neural networks to extract principal components by stabilizing synaptic weight growth.
Statements (47)
| Predicate | Object |
|---|---|
| instanceOf |
learning rule
ⓘ
synaptic plasticity rule ⓘ unsupervised learning algorithm ⓘ |
| appliesTo |
linear feedforward networks
ⓘ
single linear neuron ⓘ |
| assumes |
stationary input statistics
ⓘ
zero-mean input data ⓘ |
| basedOn | Hebbian learning ⓘ |
| category |
Hebbian learning rules
ⓘ
PCA learning rules ⓘ |
| contrastsWith | standard Hebbian rule without normalization ⓘ |
| convergesTo |
first principal component
ⓘ
leading eigenvector of input covariance matrix ⓘ |
| countryOfOrigin | Finland ⓘ |
| definedIn | “Simplified neuron model as a principal component analyzer” NERFINISHED ⓘ |
| ensures | bounded weight norm ⓘ |
| extendedTo | multi-neuron PCA networks ⓘ |
| field |
computational neuroscience
NERFINISHED
ⓘ
machine learning ⓘ neural computation ⓘ |
| hasComponent |
Hebbian term
ⓘ
weight decay term ⓘ |
| hasParameter | learning rate ⓘ |
| hasProperty | normalized Hebbian learning ⓘ |
| inspired | neural PCA algorithms ⓘ |
| introducedBy | Erkki Oja NERFINISHED ⓘ |
| learningType | unsupervised ⓘ |
| mathematicallyRelatedTo |
eigenvalue problem
ⓘ
stochastic gradient ascent ⓘ |
| maximizes | output variance under unit-norm constraint ⓘ |
| normalizes | weight vector magnitude ⓘ |
| optimizes | variance of neuron output ⓘ |
| prevents | unbounded synaptic weight growth ⓘ |
| publicationYear | 1982 ⓘ |
| publishedIn | Journal of Mathematical Biology NERFINISHED ⓘ |
| relatedTo |
Kohonen learning rule
ⓘ
Sanger rule NERFINISHED ⓘ |
| stabilizes | synaptic weights ⓘ |
| updateType | online learning rule ⓘ |
| usedFor |
adaptive signal processing
ⓘ
dimensionality reduction ⓘ feature extraction ⓘ principal component analysis ⓘ principal component extraction ⓘ subspace tracking ⓘ |
| usedIn |
neural networks
NERFINISHED
ⓘ
unsupervised neural learning ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.