Recent studies have highlighted adversarial examples as a ubiquitous threat to different neural network models and many downstream applications. Nonetheless, as unique data properties have inspired distinct and powerful learning principles, this paper aims to explore their potentials towards mitigating adversarial inputs. In particular, our results reveal the importance of using the temporal dependency in audio data to gain discriminate power against adversarial examples. Tested on the automatic speech recognition (ASR) tasks and three recent audio adversarial attacks, we find that (i) input transformation developed from image adversarial defense provides limited robustness improvement and is subtle to advanced attacks; (ii) temporal dependency can be exploited to gain discriminative power against audio adversarial examples and is resistant to adaptive attacks considered in our experiments. Our results not only show promising means of improving the robustness of ASR systems, but also offer novel insights in exploiting domain-specific data properties to mitigate negative effects of adversarial examples.
BackgroundResidual sensorimotor deficits are common following stroke. While it has been demonstrated that targeted practice can result in improvements in functional mobility years post stroke, there is little to support rehabilitation across the lifespan. The use of technology in home rehabilitation provides an avenue to better support self-management of recovery across the lifespan. We developed a novel mobile technology, capable of quantifying quality of movement with the purpose of providing feedback to augment rehabilitation and improve functional mobility. This mobile rehabilitation system, mRehab, consists of a smartphone embedded in three dimensional printed items representing functional objects found in the home. mRehab allows individuals with motor deficits to practice activities of daily living (ADLs) and receive feedback on their performance. The aim of this study was to assess the usability and consistency of measurement of the mRehab system.MethodsTo assess usability of the mRehab system, four older adults and four individuals with stroke were recruited to use the system, and complete surveys to discuss their opinions on the user interface of the smartphone app and the design of the 3D printed items. To assess the consistency of measurement by the mRehab system, 12 young adults were recruited and performed mRehab ADLs in three lab sessions within 1 week. Young adults were chosen for their expected high level of consistency in motor performance.ResultsUsability ratings from older adults and individuals with stroke led us to modify the design of the 3D printed items and improve the clarity of the mRehab app. The modified mRehab system was assessed for consistency of measurement and six ADLs resulted in coefficient of variation (CV) below 10%. This is a commonly used CV goal for consistency. Two ADLs ranged between 10 and 15% CV. Only two ADLs demonstrated high CV.ConclusionsmRehab is a client-centered technology designed for home rehabilitation that consistently measures performance. Development of the mRehab system provides a support for individuals working on recovering functional upper limb mobility that they can use across their lifespan.
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.