Multistructure-Based Collaborative Online Distillation.

Recently, deep learning has achieved state-of-the-art performance in more aspects than traditional shallow architecture-based machine-learning methods. However, in order to achieve higher accuracy, it is usually necessary to extend the network depth or ensemble the results of different neural networks. Increasing network depth or ensembling different networks increases the demand for memory resources and computing resources. This leads to difficulties in deploying depth-learning models in resource-constrained scenarios such as drones, mobile phones, and autonomous driving. Improving network performance without expanding the network scale has become a hot topic for research. In this paper, we propose a cross-architecture online-distillation approach to solve this problem by transmitting supplementary information on different networks. We use the ensemble method to aggregate networks of different structures, thus forming better teachers than traditional distillation methods. In addition, discontinuous distillation with progressively enhanced constraints is used to replace fixed distillation in order to reduce loss of information diversity in the distillation process. Our training method improves the distillation effect and achieves strong network-performance improvement. We used some popular models to validate the results. On the CIFAR100 dataset, AlexNet's accuracy was improved by 5.94%, VGG by 2.88%, ResNet by 5.07%, and DenseNet by 1.28%. Extensive experiments were conducted to demonstrate the effectiveness of the proposed method. On the CIFAR10, CIFAR100, and ImageNet datasets, we observed significant improvements over traditional knowledge distillation.

Analysis of counts with two latent classes, with application to risk assessment based on physician-visit records of cancer survivors.

Motivated by a cancer survivorship program, this paper explores event counts from two categories of individuals with unobservable membership. We formulate the counts using a latent class model and consider two likelihood-based inference procedures, the maximum likelihood estimation (MLE) and a pseudo-MLE procedure. The pseudo-MLE utilizes additional information on one of the latent classes. It yields reduced computational intensity and potentially increased estimation efficiency. We establish the consistency and asymptotic normality of the proposed pseudo-MLE, and we present an extended Huber sandwich estimator as a robust variance estimator for the pseudo-MLE. The finite-sample properties of the two-parameter estimators along with their variance estimators are examined by simulation. The proposed methodology is illustrated by physician-claim data from the cancer program.

Increased sampling of both genes and taxa improves resolution of phylogenetic relationships within Magnoliidae, a large and early-diverging clade of angiosperms.

Magnoliidae have been supported as a clade in the majority of large-scale molecular phylogenetic studies of angiosperms. This group consists of about 10,000 species assigned to 20 families and four orders, Canellales, Piperales, Laurales, and Magnoliales. Some relationships among the families are still largely debated. Here, we reconstruct the phylogenetic relationships of Magnoliidae as a whole, sampling 199 species (representing ca. 75% of genera) and 12 molecular markers from the three genomes (plastid atpB, matK, trnL intron, trnL-trnF spacer, ndhF, rbcL; mitochondrial atp1, matR, mtSSU, mtLSU; nuclear 18s rDNA, 26S rDNA). Maximum likelihood, Bayesian and maximum parsimony analyses yielded congruent trees, with good resolution and high support values for higher-level relationships. This study further confirms, with greater levels of support, two major clades in Magnoliidae: Canellales+Piperales and Laurales+Magnoliales. Relationships among the 20 families are, in general, well resolved and supported. Several previously ambiguous relationships are now well supported. For instance, the Aristolochiaceae s.l. (incl. Asaroideae, Aristolochioideae, and Lactoris) are monophyletic with high support when Hydnoraceae are excluded. The latter family was not included in most previous studies because of the lack of suitable plastid sequences, a consequence of the parasitic habit of its species. Here, we confirm that it belongs in Aristolochiaceae. Our analyses also provide moderate support for a sister group relationship between Lauraceae and Monimiaceae. Conversely, the exact position of Magnoliaceae remains very difficult to determine. This study provides a robust phylogenetic background to address the evolutionary history of an important and highly diverse clade of early-diverging angiosperms.

A sensitivity analysis using alternative toxic equivalency factors to estimate U.S. dietary exposures to dioxin-like compounds.

EPA recommends sensitivity analyses when applying the toxic equivalency factor (TEF) method to evaluate exposures to dioxin-like compounds (DLCs). Applying the World Health Organization's (WHO) 2005 TEF values and estimating average U.S. daily dietary intakes of 25 DLCs from eight food categories, we estimate a toxic equivalency (TEQ) intake of 23 pg/day. Among DLCs, PCB 126 (26%) and 1,2,3,7,8-PeCDD (23%) dominate TEQ intakes. Among food categories, milk (14%), other dairy (28%), beef (25%), and seafood (18%) most influenced TEQ intakes. We develop two approaches to estimate alternative TEF values. Based on WHO's assumption regarding TEF uncertainty, Approach1 estimates upper and lower TEFs for each DLC by multiplying and dividing, respectively, its individual TEF by ± half a log. Based on compiled empirical ranges of relative potency estimates, Approach2 uses percentile values for individual TEFs. Total TEQ intake estimates using the lower and upper TEFs based on Approach1 were 8 and 68 pg TEQ/day, respectively. The 25th and 75th percentile TEFs from Approach2 yielded 12 and 28 pg TEQ/day, respectively. The influential DLCs and food categories remained consistent across alternative TEFs, except at the 90th percentile using Approach2. We highlight the need for developing underlying TEF probability distributions.

When every particle matters: A QuEChERS approach to extract microplastics from environmental samples.

The identification of microplastics (MP), especially small (<500 µm) MP, using automated surface-chemistry approaches requires the best possible reduction of natural particles whilst preserving the integrity of the targeted synthetic polymers particles. In general, both natural and synthetic particles can be highly diverse physically and chemically and MP extraction, particularly from complex matrices such as sediments, sludge and soils, requires efficient method pipelines. Our paper presents a universal framework of modular protocols (presented in a decision tree) that fulfil predefined user requirements (QuEChERS: Quick, Easy, Cheap, Effective, Rugged, Safe) as well as providing best practises for reasonable MP working conditions within a standard laboratory. New procedures and technical innovations for density separation of particle-rich matrices are presented, such as a spiral conveyor developed and validated for MP recovery. In sharing such best-practice protocols, we aim to help in the push towards MP quantification method standardisation. •Publication of protocols of an entire MP extraction (10 µm - 5 mm) pipeline for particle-based analysis of various environmental matrices•Modularity: Optimised quantitative sample preparation adapted to particle sizes and sample matrices•New protocols and technical innovations (e.g. spiral conveyor) optimise MP extraction.

Longitudinal data on cortical thickness before and after working memory training.

The data and supplementary information provided in this article relate to our research article "Task complexity and location specific changes of cortical thickness in executive and salience networks after working memory training" (Metzler-Baddeley et al., 2016) [1]. We provide cortical thickness and subcortical volume data derived from parieto-frontal cortical regions and the basal ganglia with the FreeSurfer longitudinal analyses stream (http://surfer.nmr.mgh.harvard.edu [2]) before and after Cogmed working memory training (Cogmed and Cogmed Working Memory Training, 2012) [3]. This article also provides supplementary information to the research article, i.e., within-group comparisons between baseline and outcome cortical thickness and subcortical volume measures, between-group tests of performance changes in cognitive benchmark tests (www.cambridgebrainsciences.com [4]), correlation analyses between performance changes in benchmark tests and training-related structural changes, correlation analyses between the time spent training and structural changes, a scatterplot of the relationship between cortical thickness measures derived from the occipital lobe as control region and the chronological order of the MRI sessions to assess potential scanner drift effects and a post-hoc vertex-wise whole brain analysis with FreeSurfer Qdec (https://surfer.nmr.mgh.harvard.edu/fswiki/Qdec [5]).

Altered neural connectivity during response inhibition in adolescents with attention-deficit/hyperactivity disorder and their unaffected siblings.

INTRODUCTION: Response inhibition is one of the executive functions impaired in attention-deficit/hyperactivity disorder (ADHD). Increasing evidence indicates that altered functional and structural neural connectivity are part of the neurobiological basis of ADHD. Here, we investigated if adolescents with ADHD show altered functional connectivity during response inhibition compared to their unaffected siblings and healthy controls. METHODS: Response inhibition was assessed using the stop signal paradigm. Functional connectivity was assessed using psycho-physiological interaction analyses applied to BOLD time courses from seed regions within inferior- and superior frontal nodes of the response inhibition network. Resulting networks were compared between adolescents with ADHD (N = 185), their unaffected siblings (N = 111), and controls (N = 125). RESULTS: Control subjects showed stronger functional connectivity than the other two groups within the response inhibition network, while subjects with ADHD showed relatively stronger connectivity between default mode network (DMN) nodes. Stronger connectivity within the response inhibition network was correlated with lower ADHD severity, while stronger connectivity with the DMN was correlated with increased ADHD severity. Siblings showed connectivity patterns similar to controls during successful inhibition and to ADHD subjects during failed inhibition. Additionally, siblings showed decreased connectivity with the primary motor areas as compared to both participants with ADHD and controls. DISCUSSION: Subjects with ADHD fail to integrate activation within the response inhibition network and to inhibit connectivity with task-irrelevant regions. Unaffected siblings show similar alterations only during failed stop trials, as well as unique suppression of motor areas, suggesting compensatory strategies. These findings support the role of altered functional connectivity in understanding the neurobiology and familial transmission of ADHD.