Though recent research has achieved remarkable progress in generating realistic images with generative adversarial networks (GANs), the lack of training stability is still a lingering concern of most GANs, especially on high-resolution inputs and complex datasets. Since the randomly generated distribution can hardly overlap with the real distribution, training GANs often suffers from the gradient vanishing problem. A number of approaches have been proposed to address this issue by constraining the discriminator's capabilities using empirical techniques, like weight clipping, gradient penalty, spectral normalization etc. In this paper, we provide a more principled approach as an alternative solution to this issue. Instead of training the discriminator to distinguish real and fake input samples, we investigate the relationship between paired samples by training the discriminator to separate paired samples from the same distribution and those from different distributions. To this end, we explore a relation network architecture for the discriminator and design a triplet loss which performs better generalization and stability. Extensive experiments on benchmark datasets show that the proposed relation discriminator and new loss can provide significant improvement on variable vision tasks including unconditional and conditional image generation and image translation.
Describes an approach to normalizing microarray expression data. The novel feature is to unify the tasks of estimating normalization coefficients and identifying the control gene set. Unification is realized by constructing a window function over the scatter plot defining the subset of constantly expressed genes and by affecting optimization using an iterative procedure. The structure of window function gates contributions to the control gene set used to estimate normalization coefficients. This window measures the consistency of the matched neighborhoods in the scatter plot and provides a means of rejecting control gene outliers. The recovery of normalizational regression and control gene selection are interleaved and are realized by applying coupled operations to the mean square error function. In this way, the two processes bootstrap one another. We evaluate the technique on real microarray data from breast cancer cell lines and complement the experiment with a data cluster visualization study.
Spotted cDNA microarrays are emerging as a cost effective tool for the large scale analysis of gene expression. To reveal the patterns of genes expressed within a specific cell essentially responsible for its phenotype, this paper reports our progress in cluster discovery using a newly developed data mining method. The discussion entails: (1) statistical modeling of gene microarray data with a standard finite normal mixture distribution, (2) development of a joint supervised and unsupervised discriminative mining to discover sample clusters in a visual pyramid, and (3) evaluation of the data clusters produced by such scheme with phenotype-known microarray experiments.
Abstract Background Age of onset(AOO) influences the prognosis of many diseases and even serves as potential driver. But in Alcohol Use Disorders(AUD), there is no consensus regarding the effect of AOO on the course. Objectives We thus retrospectively investigated a large-scale cohort to explore the effect of AOO on the course of AUD. Method In this study, a total of 14,357 inpatients with alcohol use disorders were recruited over a 7- year time span, and after applying rigorous data criteria, 2,176 inpatients were ultimately included in the statistical analysis. Patients were divided into three age subgroups according to their age of onset, which were early adult onset(EAO), middle adult onset(MAO) and late adult onset(LAO). Results The proportion of recurrence was statistically different in the subgroups with different age of onset (X2=9.819, df=3, P=0.007), the Bonferroni post hoc test suggested a higher proportion of recurrence in the EAO than that in the MAO subgroup (66.1% vs 59.5%, P<0.05). We then performed a Binary logistic regression. Taking patients in the EAO subgroup as a reference, patients with MAO had a lower risk of recurrence (OR=0.75, 95%CI=0.63-0.90, P=0.002), however, the risk of recurrence in the LAO subgroup was not statistically different from the EAO (OR=0.78, 95%CI=0.54-1.11, P=0.17). Subsequently, a survival analysis for recurrence within one year was performed. There were statistically differences in Kaplan-Meier method estimates of the probability of disease-free survival (DFS) at day 365 between the three age of onset subgroups (33.9%, 40.5%, 39.7%, respectively; P=0.03). Specifically, post hoc tests suggested that DFS was lower in the EAO subgroup than in the other two groups (median DFS=197, 95%CI=172.10-221.90, P=0.012). Finally, we follow-up all patients for 5 years for cerebral atrophy onset, and the results suggest that the incidence of cerebral atrophy during the follow-up period was lower in the EAO subgroup than that in the other two subgroups (OR=1.42, 95%CI=1.19-1.70, P<0.001; OR=1.62, 95%CI=1.14-2.31, P=0.007, respectively), but this difference was no longer statistically significant after controlling for the age factor (both P>0.05). Conclusion This is the first large sample size study to explore the influence of age of onset on alcohol use disorders in a Chinese population. Our study found that alcohol use disorders with onset in early adult are associated with more recurrences. These findings will help explore risk factors for alcohol use disorders and provide preliminary evidence for clinical development of prevention strategies.
The prediction of academic award winners is a challenging task because of the complexities of establishing the appropriate evaluation system and the difficulties in modeling. Although some scholars have studied the prediction of some award winners, the related research on the prediction mechanism of academic awards in the computer field using the academic network and neural network model is still lacking. Therefore, we propose the Co-LNP mechanism, which is an academic award prediction mechanism based on a scholar cooperation network towards NSFDYS. We design a prediction model named CEw-LSTM model and provide the "improved LSTM model + artificial rule" prediction strategy to improve the prediction performance and effectiveness. Empirical studies demonstrate that the Co-LNP academic award prediction mechanism proposed in this paper has a good effect and practical application value in the computer field.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)
