Résultats de recherche pour « Robustness »

Human visual systems are robust to a wide range of image transformations that are challenging for artificial networks. We present the first study of image model robustness to the minute transformations found across video frames, which we term "natural robustness". Compared to previous studies on adversarial examples and synthetic distortions, natural robustness captures a more diverse set of common image transformations that occur in the natural environment. Our study across a dozen model architectures shows that more accurate models are more robust to natural transformations, and that robustness to synthetic color distortions is a good proxy for natural robustness. In examining brittleness in videos, we find that majority of the brittleness found in videos lies outside the typical definition of adversarial examples (99.9\%). Finally, we investigate training techniques to reduce brittleness and find that no single technique systematically improves natural robustness across twelve tested architectures.

We present a new approach for estimating parameters in rational ODE models from given (measured) time series data. In typical existing approaches, an initial guess for the parameter values is made from a given search interval. Then, in a loop, the corresponding outputs are computed by solving the ODE numerically, followed by computing the error from the given time series data. If the error is small, the loop terminates and the parameter values are returned. Otherwise, heuristics/theories are used to possibly improve the guess and continue the loop. These approaches tend to be non-robust in the sense that their accuracy depend on the search interval and the true parameter values; furthermore, they cannot handle the case where the parameters are locally identifiable. In this paper, we propose a new approach, which does not suffer from the above non-robustness. In particular, it does not require making good initial guesses for the parameter values or specifying search intervals. Instead, it uses differential algebra, interpolation of the data using rational functions, and multivariate polynomial system solving. We also compare the performance of the resulting software with several other estimation software packages.

Interpretability is an emerging area of research in trustworthy machine learning. Safe deployment of machine learning system mandates that the prediction and its explanation be reliable and robust. Recently, it has been shown that the explanations could be manipulated easily by adding visually imperceptible perturbations to the input while keeping the model's prediction intact. In this work, we study the problem of attributional robustness (i.e. models having robust explanations) by showing an upper bound for attributional vulnerability in terms of spatial correlation between the input image and its explanation map. We propose a training methodology that learns robust features by minimizing this upper bound using soft-margin triplet loss. Our methodology of robust attribution training (\textit{ART}) achieves the new state-of-the-art attributional robustness measure by a margin of $\approx$ 6-18 $\%$ on several standard datasets, ie. SVHN, CIFAR-10 and GTSRB. We further show the utility of the proposed robust training technique (\textit{ART}) in the downstream task of weakly supervised object localization by achieving the new state-of-the-art performance on CUB-200 dataset.

Highly optimized tolerance (HOT) was recently introduced as a conceptual framework to study fundamental aspects of complexity. HOT is motivated primarily by systems from biology and engineering and emphasizes, (i) highly structured, nongeneric, self-dissimilar internal configurations, and (ii) robust yet fragile external behavior. HOT claims these are the most important features of complexity and not accidents of evolution or artifices of engineering design but are inevitably intertwined and mutually reinforcing. In the spirit of this collection, our paper contrasts HOT with alternative perspectives on complexity, drawing on real-world examples and also model systems, particularly those from self-organized criticality.

While the relative trade-offs between sparse and distributed representations in deep neural networks (DNNs) are well-studied, less is known about how these trade-offs apply to representations of semantically-meaningful information. Class selectivity, the variability of a unit's responses across data classes or dimensions, is one way of quantifying the sparsity of semantic representations. Given recent evidence showing that class selectivity can impair generalization, we sought to investigate whether it also confers robustness (or vulnerability) to perturbations of input data. We found that mean class selectivity predicts vulnerability to naturalistic corruptions; networks regularized to have lower levels of class selectivity are more robust to corruption, while networks with higher class selectivity are more vulnerable to corruption, as measured using Tiny ImageNetC and CIFAR10C. In contrast, we found that class selectivity increases robustness to multiple types of gradient-based adversarial attacks. To examine this difference, we studied the dimensionality of the change in the representation due to perturbation, finding that decreasing class selectivity increases the dimensionality of this change for both corruption types, but with a notably larger increase for adversarial attacks. These results demonstrate the causal relationship between selectivity and robustness and provide new insights into the mechanisms of this relationship.

In this paper we establish rigorous benchmarks for image classifier robustness. Our first benchmark, ImageNet-C, standardizes and expands the corruption robustness topic, while showing which classifiers are preferable in safety-critical applications. Unlike recent robustness research, this benchmark evaluates performance on commonplace corruptions not worst-case adversarial corruptions. We find that there are negligible changes in relative corruption robustness from AlexNet to ResNet classifiers, and we discover ways to enhance corruption robustness. Then we propose a new dataset called Icons-50 which opens research on a new kind of robustness, surface variation robustness. With this dataset we evaluate the frailty of classifiers on new styles of known objects and unexpected instances of known classes. We also demonstrate two methods that improve surface variation robustness. Together our benchmarks may aid future work toward networks that learn fundamental class structure and also robustly generalize.

Biological systems, from macromolecules to whole organisms, are robust if they continue to function, survive, or reproduce when faced with mutations, environmental change, and internal noise. I focus here on biological systems that are robust to mutations and ask whether such systems are more or less evolvable, in the sense that they can acquire novel properties. The more robust a system is, the more mutations in it are neutral, that is, without phenotypic effect. I argue here that such neutral change--and thus robustness--can be a key to future evolutionary innovation, if one accepts that neutrality is not an essential feature of a mutation. That is, a once neutral mutation may cause phenotypic effects in a changed environment or genetic background. I argue that most, if not all, neutral mutations are of this sort, and that the essentialist notion of neutrality should be abandoned. This perspective reconciles two opposing views on the forces dominating organismal evolution, natural selection and random drift: neutral mutations occur and are especially abundant in robust systems, but they do not remain neutral indefinitely, and eventually become visible to natural selection, where some of them lead to evolutionary innovations.