No abstract
This paper investigates strategies that defend against adversarial-example attacks on image-classification systems by transforming the inputs before feeding them to the system. Specifically, we study applying image transformations such as bit-depth reduction, JPEG compression, total variance minimization, and image quilting before feeding the image to a convolutional network classifier. Our experiments on ImageNet show that total variance minimization and image quilting are very effective defenses in practice, in particular, when the network is trained on transformed images. The strength of those defenses lies in their non-differentiable nature and their inherent randomness, which makes it difficult for an adversary to circumvent the defenses. Our best defense eliminates 60% of strong gray-box and 90% of strong black-box attacks by a variety of major attack methods.
Generating adversarial examples is a critical step for evaluating and improving the robustness of learning machines. So far, most existing methods only work for classification and are not designed to alter the true performance measure of the problem at hand. We introduce a novel flexible approach named Houdini for generating adversarial examples specifically tailored for the final performance measure of the task considered, be it combinatorial and non-decomposable. We successfully apply Houdini to a range of applications such as speech recognition, pose estimation and semantic segmentation. In all cases, the attacks based on Houdini achieve higher success rate than those based on the traditional surrogates used to train the models while using a less perceptible adversarial perturbation.
ConvNets and ImageNet have driven the recent success of deep learning for image classification. However, the marked slowdown in performance improvement combined with the lack of robustness of neural networks to adversarial examples and their tendency to exhibit undesirable biases question the reliability of these methods. This work investigates these questions from the perspective of the end-user by using human subject studies and explanations. The contribution of this study is threefold. We first experimentally demonstrate that the accuracy and robustness of ConvNets measured on ImageNet are vastly underestimated. Next, we show that explanations can mitigate the impact of misclassified adversarial examples from the perspective of the end-user. We finally introduce a novel tool for uncovering the undesirable biases learned by a model. These contributions also show that explanations are a valuable tool both for improving our understanding of ConvNets' predictions and for designing more reliable models.
Background/methodsInsecticide-treated nets (ITNs) are the primary tool for malaria vector control in sub-Saharan Africa, and have been responsible for an estimated two-thirds of the reduction in the global burden of malaria in recent years. While the ultimate goal is high levels of ITN use to confer protection against infected mosquitoes, it is widely accepted that ITN use must be understood in the context of ITN availability. However, despite nearly a decade of universal coverage campaigns, no country has achieved a measured level of 80% of households owning 1 ITN for 2 people in a national survey. Eighty-six public datasets from 33 countries in sub-Saharan Africa (2005–2017) were used to explore the causes of failure to achieve universal coverage at the household level, understand the relationships between the various ITN indicators, and further define their respective programmatic utility.ResultsThe proportion of households owning 1 ITN for 2 people did not exceed 60% at the national level in any survey, except in Uganda’s 2014 Malaria Indicator Survey (MIS). At 80% population ITN access, the expected proportion of households with 1 ITN for 2 people is only 60% (p = 0.003 R2 = 0.92), because individuals in households with some but not enough ITNs are captured as having access, but the household does not qualify as having 1 ITN for 2 people. Among households with 7–9 people, mean population ITN access was 41.0% (95% CI 36.5–45.6), whereas only 6.2% (95% CI 4.0–8.3) of these same households owned at least 1 ITN for 2 people. On average, 60% of the individual protection measured by the population access indicator is obscured when focus is put on the household “universal coverage” indicator. The practice of limiting households to a maximum number of ITNs in mass campaigns severely restricts the ability of large households to obtain enough ITNs for their entire family.ConclusionsThe two household-level indicators—one representing minimal coverage, the other only ‘universal’ coverage—provide an incomplete and potentially misleading picture of personal protection and the success of an ITN distribution programme. Under current ITN distribution strategies, the global malaria community cannot expect countries to reach 80% of households owning 1 ITN for 2 people at a national level. When programmes assess the success of ITN distribution activities, population access to ITNs should be considered as the better indicator of “universal coverage,” because it is based on people as the unit of analysis.Electronic supplementary materialThe online version of this article (10.1186/s12936-018-2505-0) contains supplementary material, which is available to authorized users.
No abstract
BackgroundEffective case management of malaria requires prompt diagnosis and treatment within 24 hours. Home-based management of malaria (HMM) improves access to treatment for populations with limited access to health facilities. In Senegal, an HMM pilot study in 2008 demonstrated the feasibility of integrated use of RDTs and ACT in remote villages by volunteer Home Care Providers (HCP). Scale-up of the strategy began in 2009, reaching 408 villages in 2009 and 861 villages in 2010. This paper reports the results of the scale-up in the targeted communities and the impact of the strategy on malaria in the formal health sector.MethodsData reported by the HCPs were used to assess their performance in 2009 and 2010, while routine malaria morbidity and mortality data were used to assess the impact of the HMM programme. Two high transmission regions where HMM was not implemented until 2010 were used as a comparison.Results and discussionFrom July 2009 through May 2010, 12582 suspected cases were managed by HCPs, 93% (11672) of whom were tested with an RDT. Among those tested, 37% (4270) had a positive RDT, 97% (4126) of whom were reported treated and cured. Home care providers referred 6871 patients to health posts for management: 6486 with a negative RDT, 119 infants < 2 months, 105 pregnant women, and 161 severe cases. There were no deaths among these patients. In 2009 compared to 2008, incidence of suspected and confirmed malaria cases, all hospitalizations and malaria-related hospitalizations decreased in both intervention and comparison regions. Incidence of in-hospital deaths due to malaria decreased by 62.5% (95% CI 43.8-81.2) in the intervention regions, while the decrease in comparison regions was smaller and not statistically significant.ConclusionHome-based management of malaria including diagnosis with RDT and treatment based on test results is a promising strategy to improve the access of remote populations to prompt and effective management of uncomplicated malaria and to decrease mortality due to malaria. When scaled-up to serve remote village communities in the regions of Senegal with the highest malaria prevalence, home care providers demonstrated excellent adherence to guidelines, potentially contributing to a decrease in hospital deaths attributed to malaria.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.