boundlab.poly.PolyBounds#

class boundlab.poly.PolyBounds[source]#

Bases: object

CROWN-style linear relaxation bounds for a unary nonlinearity.

Represents pointwise constraints

\[\lambda_\ell \odot x + b_\ell \;\le\; f(x) \;\le\; \lambda_u \odot x + b_u.\]

The constituent tensors are per-neuron and share the activation’s input/output shape.

Methods

upper_lam: torch.Tensor#
upper_bias: torch.Tensor#
lower_lam: torch.Tensor#
lower_bias: torch.Tensor#
__init__(upper_lam, upper_bias, lower_lam, lower_bias)#