dopt.nnet.lipschitz

Contains an implementation of the regularisation techniques presented in Gouk et al. (2018).

Gouk, H., Frank, E., Pfahringer, B., & Cree, M. (2018). Regularisation of Neural Networks by Enforcing Lipschitz Continuity. arXiv preprint arXiv:1804.04368.

Members

Functions

convParamsNorm
Operation convParamsNorm(Operation param, size_t[] inShape, size_t[] stride, size_t[] padding, float p, size_t n)
Undocumented in source. Be warned that the author may not have intended to support it.
matrixNorm
Operation matrixNorm(Operation param, float p)

Computes the induced operator norm corresponding to the vector p-norm.

maxNorm
Operation maxNorm(Operation param, Operation norm, Operation maxval)

Performs a projection of param such that the new norm will be less than or equal to maxval.

projConvParams
Projection projConvParams(Operation maxnorm, size_t[] inShape, size_t[] stride, size_t[] padding, float p)
Undocumented in source. Be warned that the author may not have intended to support it.
projMatrix
Projection projMatrix(Operation maxnorm, float p)

Returns a projection function that can be used to constrain a matrix norm.

Meta

Authors

Henry Gouk