Search

Algorithmic evaluation is a vital step in developing new approaches to machine learning and relies on the availability of existing datasets. However, real-world datasets often do not cover the necessary complexity space required to understand an algorithm's domains of competence. As such, the generation of synthetic datasets to fill gaps in the complexity space has gained attention, offering a means of evaluating algorithms when data is unavailable. Existing approaches to complexity-focused data generation are limited in their ability to generate solutions that invoke similar classification behaviour to real data. The present work proposes a novel method (Sy:Boid) for complexity-based synthetic data generation, adapting and extending the Boid algorithm that was originally intended for computer graphics simulations. Sy:Boid embeds the modified Boid algorithm within an evolutionary multi-objective optimisation algorithm to generate synthetic datasets which satisfy predefined magnitudes of complexity measures. Sy:Boid is evaluated and compared to labelling-based and sampling-based approaches to data generation to understand its ability to generate a wide variety of realistic datasets. Results demonstrate Sy:Boid is capable of generating datasets across a greater portion of the complexity space than existing approaches. Furthermore, the produced datasets were observed to invoke very similar classification behaviours to that of real data. [ABSTRACT FROM AUTHOR]

Human health is threatened by several diseases for this reason automated medical diagnosis systems has been developed several years ago. These systems need databases, the creation of these databases is tedious, arduous and stops being done so the created database is incomplete or unbalanced. Sometimes the databases are private to protect the private information of the patients, among other problems. For this reason, the researchers have started to use synthetic data. The synthetic data have been applied by different hospitals in the USA. The creation of synthetic data has different problems like the synthetic data are generated using rules defined by the user, the proposed approaches only can create one kind of data, the proposals require input from domain experts, among others. To address these kinds of problems, we propose a novel approach, which consists of the Bidirectional Recurrent Neural Network and the statistical stage to generate synthetic biomedical signals. The approach is able to create 5 kinds of biomedical signals (ECG, EEG, BCG, PPG, and Respiratory Impedance). Our approach is able to create synthetic data for patients or for specific events. The performance of our approach is compared with other generative models (GAN's) through evaluation metrics. The created synthetic data are used to construct models; these models are able to successfully differentiate between different signals with high accuracies. [ABSTRACT FROM AUTHOR]

Synthetic data describing coherent random fluctuations have widely been used to validate numerical simulations against experimental observations or to examine the reliability of extracting statistical properties of plasma turbulence via correlation functions. Estimating correlation time or lengths based on correlation functions implicitly assumes that the observed data are stationary and homogeneous . It is, therefore, important that numerically generated synthetic data also satisfy the stationary process and homogeneous state. Based on the synthetic data with randomly generated moving Gaussian shaped fluctuations both in time and space, the correlation function depending on the size of averaging time window is analytically derived. Then, the smallest possible spatial window size of synthetic data satisfying the stationary process and homogeneous state is proposed, thereby reducing the computation time to generate proper synthetic data and providing a constraint on the minimum size of simulation domains when using synthetic diagnostics to compare with experiment. This window size is also numerically confirmed with 1D synthetic data with various parameter scans. [ABSTRACT FROM AUTHOR]

In recent years, deep learning has demonstrated tremendous potential in the real world. Object detection is a critical real-world task for deep learning. You Only Look Once (YOLO) object detection model recognizes interesting regions in images with impressive accuracy and real-time performance. The objective of this paper is to apply object detection to the field of security and counter-terrorism. Individuals are protected from violence by recognizing and locating the guns on closed-circuit television (CCTV). This paper presents a real-time detection approach for CCTV autonomous weapons based on YOLO v4. For the characteristics of CCTV scenarios, we propose the YOLO v4 backbone with Spatial Cross Stage Partial-ResNet (SCSP-ResNet). Meanwhile, the receptive field enhancement module is shown to capture fine semantic features of high-dimensional small objects. The Fusion-PANet (F-PaNet) module has been used to fuse multi-scale information to improve the model's perceptive power on the region of interest. Furthermore, we merge synthetic and real-world datasets to comprehensively investigate the effects of synthetic datasets on detectors. The experimental results reveal that our suggested detection model improves mAP (mean Accuracy Precision) and inference time by 7.37% and 4.2%, respectively. The model's parameter is reduced by 0.349 BFLOP/s(billion floating point operations per second). The proposed detector outperforms the baseline model in terms of accuracy, real-time, and robustness. [ABSTRACT FROM AUTHOR]

The partially linear single-index model is an interesting semiparametric model extended by the partially linear model and the single-index model, which supply a good balance between flexibility and parsimony. A robust estimation is proposed to fit the partially linear single-index model in case outliers may occur in the right censored response. This method provides a flexible way for modeling survival data. It is a profile M-estimation version and the estimation procedure involves transforming the censored data into synthetic data at first, then it results in fitting the common partially linear single-index models by a robust loss function. Asymptotic properties for the estimators of the linear and single-index coefficients and the optimal rate of convergence for the estimator of the nonparametric function are established. The finite sample performance of the proposed method is assessed by Monte Carlo simulation studies, and demonstrated by the analyses of PBC data and NCCTG lung cancer data. [ABSTRACT FROM AUTHOR]

Abstract: Statistical disclosure control (also known as privacy-preserving data mining) of microdata is about releasing data sets containing the answers of individual respondents protected in such a way that: (i) the respondents corresponding to the released records cannot be re-identified; (ii) the released data stay analytically useful. Usually, the protected data set is generated by either masking (i.e. perturbing) the original data or by generating synthetic (i.e. simulated) data preserving some pre-selected statistics of the original data. Masked data may approximately preserve a broad range of distributional characteristics, although very few of them (if any) are exactly preserved; on the other hand, synthetic data exactly preserve the pre-selected statistics and may seem less disclosive than masked data, but they do not preserve at all any statistics other than those pre-selected. Hybrid data obtained by mixing the original data and synthetic data have been proposed in the literature to combine the strengths of masked and synthetic data. We show how to easily obtain hybrid data by combining microaggregation with any synthetic data generator. We show that numerical hybrid data exactly preserving means and covariances of original data and approximately preserving other statistics as well as some subdomain analyses can be obtained as a particular case with a very simple parameterization. The new method is competitive versus both the literature on hybrid data and plain multivariate microaggregation. [Copyright &y& Elsevier]

Abstract: Multiple imputation can handle missing data and disclosure limitation simultaneously. First, fill in the missing data to generate m completed datasets, then replace confidential values in each completed dataset with r imputations. I investigate how to select m and r. [Copyright &y& Elsevier]

Abstract: The maximum likelihood estimators (MLEs) and the moment estimators of a two-parameter Birnbaum–Saunders (BISA) distribution are studied by various authors when data are either complete or subject to Type-I or Type-II censoring. But there is not much research on parameter estimation for the BISA distribution under random censoring. A simple method of modified censored moment estimation is proposed to estimate parameters of the BISA distribution under random censoring. Bias-reduced versions of these estimators are constructed as well. Asymptotic theory for the estimators is established. The performance of these estimators is compared with that of the MLEs through Monte Carlo simulations for small, moderate, and large proportions of censoring and different sample sizes. An analysis of real data is used to illustrate the proposed method. [Copyright &y& Elsevier]