boundlab.zono.ZonoBounds

class boundlab.zono.ZonoBounds

Bases: object

Data class representing zonotope bounds for a neural network layer.

Examples

input_weights has one entry per input expression to the linearizer. For unary ops such as ReLU, this is typically a single slope tensor.

Methods

__init__
apply_without_error	Apply the zonotope bounds to given input tensors, ignoring the error term.

bias: torch.Tensor

error_coeffs: LinearOp

input_weights: list[torch.Tensor | 0]

apply_without_error(*inputs)

Apply the zonotope bounds to given input tensors, ignoring the error term.

__init__(bias, error_coeffs, input_weights)