-
401
Wasserstein distributional robustness of neural networks
Published 2023“…Deep neural networks are known to be vulnerable to adversarial attacks (AA).For an image recognition task, this means that a small perturbation of the original can result in the image being misclassified.Design of such attacks as well as methods of adversarial training against them are subject of intense research.We re-cast the problem using techniques of Wasserstein distributionally robust optimization (DRO) and obtain novel contributions leveraging recent insights from DRO sensitivity analysis.We consider a set of distributional threat models.Unlike the traditional pointwise attacks, which assume a uniform bound on perturbation of each input data point, distributional threat models allow attackers to perturb inputs in a nonuniform way.We link these more general attacks with questions of out-of-sample performance and Knightian uncertainty.To evaluate the distributional robustness of neural networks, we propose a first-order AA algorithm and its multistep version.Our attack algorithms include Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) as special cases.Furthermore, we provide a new asymptotic estimate of the adversarial accuracy against distributional threat models.The bound is fast to compute and first-order accurate, offering new insights even for the pointwise AA.It also naturally yields out-of-sample performance guarantees.We conduct numerical experiments on CIFAR-10, CIFAR-100, ImageNet datasets using DNNs on RobustBench to illustrate our theoretical results.Our code is available at https://github.com/JanObloj/W-DRO-Adversarial-Methods.…”
Conference item -
402
Lagrangian decomposition for neural network verification
Published 2020“…A fundamental component of neural network verification is the computation of bounds on the values their outputs can take. …”
Conference item -
403
Towards efficient and reliable neural networks
Published 2024“…<p>Deep neural networks have achieved remarkable results in various applications, but when deployed to autonomous agents in the real world several challenges arise. …”
Thesis -
404
Adversarial robustness of Bayesian neural networks
Published 2021“…<p>This thesis puts forward methods for computing local robustness of probabilistic neural networks, specifically those resulting from Bayesian inference. …”
Thesis -
405
Graph neural networks for network analysis
Published 2024Subjects: “…Neural networks (computer science)…”
Thesis -
406
Formal synthesis of Lyapunov neural networks
Published 2020“…Our method supports neural networks with polynomial activation functions and multiple depth and width, which display wide learning capabilities. …”
Journal article -
407
-
408
Learning invariant representations in neural networks
Published 2021“…The presented work focuses on invariant and equivariant neural network layers, putting symmetries at the centre of neural network architecture design. …”
Thesis -
409
-
410
Probabilistic safety for bayesian neural networks
Published 2020“…We study probabilistic safety for Bayesian Neural Networks (BNNs) under adversarial input perturbations. …”
Conference item -
411
Ontology reasoning with deep neural networks
Published 2020“…In this paper, we employ state-of-the-art methods for training deep neural networks to devise a novel model that is able to learn how to effectively perform logical reasoning in the form of basic ontology reasoning. …”
Journal article -
412
-
413
-
414
-
415
Orbit-equivariant graph neural networks
Published 2024“…Equivariance is an important structural property that is captured by architectures such as graph neural networks (GNNs). However, equivariant graph functions cannot produce different outputs for similar nodes, which may be undesirable when the function is trying to optimize some global graph property. …”
Conference item -
416
Exploiting sparsity for neural network verification
Published 2021“…The problem of verifying the properties of a neural network has never been more important. This task is often done by bounding the activation functions in the network. …”
Conference item -
417
-
418
Individual fairness guarantees for neural networks
Published 2022“…We consider the problem of certifying the individual fairness (IF) of feed-forward neural networks (NNs). In particular, we work with the epsilon-delta-IF formulation, which, given a NN and a similarity metric learnt from data, requires that the output difference between any pair of epsilon-similar individuals is bounded by a maximum decision tolerance delta >= 0. …”
Conference item -
419
Encryption function on artificial neural network
Published 2016“…In this paper, we try to decrypt automatically using artificial neural network by decryption through multilayer perceptron and radial basis function; networks were tested using the interface by calculating the error rates of decrypted images.…”
Article -
420
Backpropagation Neural Network For Colour Recognition
Published 2002“…It is found that RGB is useful when used with Neural Network and the Normalized RGB value is faster in the learning of neural network.…”
Get full text
Thesis