publications

2021

1. arXiv

   Frustratingly Easy Uncertainty Estimation for Distribution Shift

   Salvador, T., Voleti, V.,  Iannantuono, A., and Oberman, A. M.

   2021

   Abs HTML PDF

   Distribution shift is an important concern in deep image classification, produced either by corruption of the source images, or a complete change, with the solution involving domain adaptation. While the primary goal is to improve accuracy under distribution shift, an important secondary goal is uncertainty estimation: evaluating the probability that the prediction of a model is correct. While improving accuracy is hard, uncertainty estimation turns out to be frustratingly easy. Prior works have appended uncertainty estimation into the model and training paradigm in various ways. Instead, we show that we can estimate uncertainty by simply exposing the original model to corrupted images, and performing simple statistical calibration on the image outputs. Our frustratingly easy methods demonstrate superior performance on a wide range of distribution shifts as well as on unsupervised domain adaptation tasks, measured through extensive experimentation.

2020

1. JSC

   No-Collision Transportation Maps

   Nurbekyan, L.,  Iannantuono, A., and Oberman, A. M.

   2020

   Abs HTML PDF

   Transportation maps between probability measures are critical objects in numerous areas of mathematics and applications such as PDE, fluid mechanics, geometry, machine learning, computer science, and economics. Given a pair of source and target measures, one searches for a map that has suitable properties and transports the source measure to the target one. Here, we study maps that possess the no-collision property; that is, particles simultaneously traveling from sources to targets in a unit time with uniform velocities do not collide. These maps are particularly relevant for applications in swarm control problems. We characterize these no-collision maps in terms of half-space preserving property and establish a direct connection between these maps and binary-space-partitioning (BSP) tree structures. Based on this characterization, we provide explicit BSP algorithms, of cost \(O(n \log n)\), to construct no-collision maps. Moreover, interpreting these maps as approximations of optimal transportation maps, we find that they succeed in computing nearly optimal maps for \(q\)-Wasserstein metric (\(q = 1, 2\)). In some cases, our maps yield costs that are just a few percent off from being optimal.

2. arXiv

   Calibrated Top-1 Uncertainty estimates for classification by score based models

   Oberman, A. M., Finlay, C.,  Iannantuono, A., and Salvador, T.

   2020

   Abs HTML PDF

   While the accuracy of modern deep learning models has significantly improved in recent years, the ability of these models to generate uncertainty estimates has not progressed to the same degree. Uncertainty methods are designed to provide an estimate of class probabilities when predicting class assignment.

2019

1. A Langevin dynamics based approach to generating sparse adversarial perturbations

   Pooladian, A-A.,  Iannantuono, A., Finlay, C., and Oberman, A. M.

   2019

   Abs PDF

   Deep neural networks are vulnerable to small changes in the input that lead to misclassification, calledadversarial images. We present an efficient approach togenerating sparse adversarial images, i.e. small with respect to the cardinality function \(\ell_0\), without using gradient information. The lack of gradient oracle isof interest in the case of adversarial attacks: in a practical setting, one can easily query the model to determine whether the image is misclassified or not but rarely will the full network structure be provided. Our method, ProxWalk, is inspired by Metropolis-Adjusted Langevin dynamics; a method for modeling random walks,and the proximal variant. We present results on MNIST, Fashion-MNIST, CIFAR10, and CIFAR100 datasets, and demonstrate that our decision-based attackis on par with modern sparse white-box attacks.