
On the Geometry of Adversarial Examples
Adversarial examples are a pervasive phenomenon of machine learning mode...
read it

Manifold Mixup: Encouraging Meaningful OnManifold Interpolation as a Regularizer
Deep networks often perform well on the data manifold on which they are ...
read it

Idealised Bayesian Neural Networks Cannot Have Adversarial Examples: Theoretical and Empirical Study
We prove that idealised discriminative Bayesian neural networks, capturi...
read it

RetrievalAugmented Convolutional Neural Networks for Improved Robustness against Adversarial Examples
We propose a retrievalaugmented convolutional network and propose to tr...
read it

Using Videos to Evaluate Image Model Robustness
Human visual systems are robust to a wide range of image transformations...
read it

The Conditional Entropy Bottleneck
Much of the field of Machine Learning exhibits a prominent set of failur...
read it

Adversarial Training with Voronoi Constraints
Adversarial examples are a pervasive phenomenon of machine learning mode...
read it
Disentangling Adversarial Robustness and Generalization
Obtaining deep networks that are robust against adversarial examples and generalize well is an open problem. A recent hypothesis even states that both robust and accurate models are impossible, i.e., adversarial robustness and generalization are conflicting goals. In an effort to clarify the relationship between robustness and generalization, we assume an underlying, lowdimensional data manifold and show that: 1. regular adversarial examples leave the manifold; 2. adversarial examples constrained to the manifold, i.e., onmanifold adversarial examples, exist; 3. onmanifold adversarial examples are generalization errors, and onmanifold adversarial training boosts generalization; 4. and regular robustness is independent of generalization. These assumptions imply that both robust and accurate models are possible. However, different models (architectures, training strategies etc.) can exhibit different robustness and generalization characteristics. To confirm our claims, we present extensive experiments on synthetic data (with access to the true manifold) as well as on EMNIST, FashionMNIST and CelebA.
READ FULL TEXT
Comments
There are no comments yet.