TorchLean API

NN.MLTheory.CROWN.Propagation.LinearSignsplit

Sign-splitting linear bounds (IBP helper) #

For a linear layer y = W x + b with input interval x ∈ [lo, hi], we can compute output bounds using the standard sign-splitting trick:

Then:

This is algebraically equivalent to the per-weight min/max rule, but it makes it easy to cache W⁺/W⁻ and matches the common LiRPA/CROWN implementation style.

Positive part of a weight matrix: W⁺ = max(W, 0) (elementwise).

Instances For

    Negative part of a weight matrix: W⁻ = min(W, 0) (elementwise).

    Instances For

      Linear IBP bounds computed via sign-splitting (W⁺/W⁻).

      Instances For