The article introduces algorithms PIE and GIE for computing tight guarantees on the probabilistic robustness of Bayesian Neural Networks (BNNs). It highlights the challenges in verifying BNNs compared to standard NNs and the limitations of existing approaches. The algorithms efficiently search for safe weights using iterative expansion and network gradients, outperforming the state-of-the-art (SoA) in terms of tightness of bounds. Evaluation on MNIST and CIFAR10 datasets shows significant improvements in bounds.
A otro idioma
del contenido fuente
arxiv.org
Ideas clave extraídas de
by Ben Batten,M... a las arxiv.org 03-01-2024
https://arxiv.org/pdf/2401.11627.pdfConsultas más profundas