Domain generalization (DG) is the challenging and topical problem of learning models that generalize to novel testing domains with different statistics than a set of known training domains. The simple approach of aggregating data from all source domains and training a single deep neural network end-to-end on all the data provides a surprisingly strong baseline that surpasses many prior published methods. In this paper we build on this strong baseline by designing an episodic training procedure that trains a single deep network in a way that exposes it to the domain shift that characterises a novel domain at runtime. Specifically, we decompose a deep network into feature extractor and classifier components, and then train each component by simulating it interacting with a partner who is badly tuned for the current domain. This makes both components more robust, ultimately leading to our networks producing state-of-the-art performance on three DG benchmarks. Furthermore, we consider the pervasive workflow of using an ImageNet trained CNN as a fixed feature extractor for downstream recognition tasks. Using the Visual Decathlon benchmark, we demonstrate that our episodic-DG training improves the performance of such a general purpose feature extractor by explicitly training a feature for robustness to novel problems. This shows that DG training can benefit standard practice in computer vision.
Background: Hospital staff are vulnerable and at high risk of novel coronavirus disease (COVID-19) infection. The aim of this study was to monitor the psychological distress in hospital staff and examine the relationship between the psychological distress and possible causes during the COVID-19 epidemic. Methods: An online survey was conducted from February 1 to February 14, 2020. Hospital staff from five national COVID-19 designated hospitals in Chongqing participated. Data collected included demographics and stress responses to COVID-19: 1) the impact of event scale to measure psychological stress reactions; 2) generalized anxiety disorder 7 to measure anxiety symptoms; 3) Patient Health Questionnaire 9 to measure depression symptoms; 4) Yale-Brown Obsessive-Compulsive Scale to measure obsessive-compulsive symptoms (OCS); and 5) Patient Health Questionnaire 15 to measure somatization symptoms. Multiple logistic regression analysis was used to identify factors that were correlated with psychological distress. Results: Hospital staff that participated in this study were identified as either doctors or nurses. A total of 456 respondents completed the questionnaires with a response rate of 91.2%. The mean age was 30.67 ± 7.48 years (range, 17 to 64 years). Of all respondents, 29.4% were men. Of the staff surveyed, 43.2% had stress reaction syndrome. The highest prevalence of psychological distress was OCS (37.5%), followed by somatization symptoms (33.3%), anxiety symptoms (31.6%), and depression symptoms (29.6%). Univariate analyses indicated that female subjects, middle aged subjects, subjects in the low income group, and subjects working in isolation wards were prone to experience psychological distress. Multiple logistic regression analysis showed “Reluctant to work or considered resignation” (odds ratio [OR], 5.192; 95%CI, 2.396–11.250; P < .001 ), “Afraid to go home because of fear of infecting family” (OR, 2.099; 95%CI, 1.299–3.391; P = .002 ) “Uncertainty about frequent modification of infection and control procedures” (OR, 1.583; 95%CI, 1.061–2.363; P = .025 ), and“Social support” (OR, 1.754; 95%CI, 1.041–2.956; P = .035 ) were correlated with psychological reactions. “Reluctant to work or considered resignation” and “Afraid to go home because of fear of infecting family” were associated with a higher risk of symptoms of Anxiety (OR, 3.622; 95% CI, 1.882–6.973; P < .001 ; OR, 1.803; 95% CI, 1.069–3.039; P = .027), OCS (OR, 5.241; 95% CI, 2.545–10.793; P < .001 ; OR, 1.999; 95% CI, 1.217–3.282; P = .006 ) and somatization (OR, 5.177; 95% CI, 2.595–10.329; P ...
Circular economy (CE) is being increasingly accepted as a promising sustainable business model, supporting waste minimization through product life cycles. The product end-of-use (EOU) stage is the key to circulate materials and components into a new life cycle, rather than direct disposal. The economic viability of recycling EOU products is significantly affected by designers' decisions and largely determined during product design. Low economic return of EOU value recovery is a major barrier to overcome. To address this issue, a design method to facilitate EOU product value recovery is proposed. First, product EOU scenarios are determined by optimization of EOU component flows. The EOU scenario depicts which modules (groups of components) will be allocated for reuse, recycling, or disposal, the order of joint detachment (the joints for modules connection), and recovery profit. Second, in the given study, bottlenecks, improvement opportunities, and design suggestions will be identified and provided following the EOU scenario analysis. Pareto analysis is used for ranking joints, according to their detachment cost and for indicating which joints are the most suitable for replacement. An analytic hierarchy process (AHP) is employed to select the best joint candidate with trade-off among criteria from the perspective of disassembly. In addition, disposal and recycling modules are checked to eliminate hazardous material and increase material compatibility. A value-based recycling indicator is developed to measure recyclability of the modules and evaluate design suggestions for material selection. Finally, based on heuristics, the most valuable and reusable modules will be selected for reconfiguration so that they can be easily accessed and disassembled. A hard disk drive is used as a case study to illustrate the method.
This paper investigates generic signal shaping methods for multiple-data-stream generalized spatial modulation (GenSM) and generalized quadrature spatial modulation (Gen-QSM) based on the maximizing the minimum Euclidean distance (MMED) criterion. Three cases with different channel state information at the transmitter (CSIT) are considered, including no CSIT, statistical CSIT and perfect CSIT. A unified optimization problem is formulated to find the optimal transmit vector set under size, power and sparsity constraints. We propose an optimization-based signal shaping (OBSS) approach by solving the formulated problem directly and a codebook-based signal shaping (CBSS) approach by finding sub-optimal solutions in discrete space. In the OBSS approach, we reformulate the original problem to optimize the signal constellations used for each transmit antenna combination (TAC). Both the size and entry of all signal constellations are optimized. Specifically, we suggest the use of a recursive design for size optimization. The entry optimization is formulated as a non-convex large-scale quadratically constrained quadratic programming (QCQP) problem and can be solved by existing optimization techniques with rather high complexity. To reduce the complexity, we propose the CBSS approach using a codebook generated by quadrature amplitude modulation (QAM) symbols and a low-complexity selection algorithm to choose the optimal transmit vector set. Simulation results show that the OBSS approach exhibits the optimal performance in comparison with existing benchmarks. However, the OBSS approach is impractical for large-size signal shaping and adaptive signal shaping with instantaneous CSIT due to the demand of high computational complexity. As a lowcomplexity approach, CBSS shows comparable performance and can be easily implemented in large-size systems.
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.