Sparse autoencoder

Source: R/autoencoder_sparse.R

Creates a representation of a sparse autoencoder.

autoencoder_sparse(network, loss = "mean_squared_error",
  high_probability = 0.1, weight = 0.2)

Arguments

network

Layer construct of class "ruta_network"
loss

Character string specifying a loss function
high_probability

Expected probability of the high value of the encoding layer. Set this to a value near zero in order to minimize activations in that layer.
weight

The weight of the sparsity regularization

Value

A construct of class "ruta_autoencoder"

References

  • Sparse deep belief net model for visual area V2

  • Andrew Ng, Sparse Autoencoder. CS294A Lecture Notes

See also

sparsity, make_sparse, is_sparse

Other autoencoder variants: autoencoder_contractive, autoencoder_denoising, autoencoder_robust, autoencoder_variational, autoencoder