Data augmentation has been proven particularly effective for image classification tasks where a significant boost of prediction accuracy can be obtained when the technique is combined with the use of Deep Learning architectures. Unfortunately, for non-image data the situation is quite different and the positive effect of augmenting the training set size is much smaller. In this work, we propose a method that creates new samples by adjusting the level of noise for individual input variables previously ranked by their relevance level. Results from several tests are analyzed using nine benchmark data sets when the augmented and original data are used for supervised training on Deep Learning architectures.

Presently, we are living in a hyper-connected world where millions of heterogeneous devices are continuously sharing information in different application contexts for wellness, improving communications, digital businesses, etc. However, the bigger the number of devices and connections are, the higher the risk of security threats in this scenario. To counteract against malicious behaviours and preserve essential security services, Network Intrusion Detection Systems (NIDSs) are the most widely used defence line in communications networks. Nevertheless, there is no standard methodology to evaluate and fairly compare NIDSs. Most of the proposals elude mentioning crucial steps regarding NIDSs validation that make their comparison hard or even impossible. This work firstly includes a comprehensive study of recent NIDSs based on machine learning approaches, concluding that almost all of them do not accomplish with what authors of this paper consider mandatory steps for a reliable comparison and evaluation of NIDSs. Secondly, a structured methodology is proposed and assessed on the UGR’16 dataset to test its suitability for addressing network attack detection problems. The guideline and steps recommended will definitively help the research community to fairly assess NIDSs, although the definitive framework is not a trivial task and, therefore, some extra effort should still be made to improve its understandability and usability further.

Type 2 diabetes results from interplay between genetic and acquired factors. Glycans on proteins reflect genetic, metabolic and environmental factors. However, associations of IgG glycans with type 2 diabetes have not been described. We compared IgG N-glycan patterns in type 2 diabetes with healthy subjects. In the DiaGene study, a population-based case-control study, (1886 cases and 854 controls) 58 IgG glycan traits were analyzed. Findings were replicated in the population-based CROATIA-Korcula-CROATIA-Vis-ORCADES studies (162 cases and 3162 controls), and meta-analyzed. AUCs of ROC-curves were calculated using 10-fold cross-validation for clinical characteristics, IgG glycans and their combination. After correction for extensive clinical covariates, 5 IgG glycans and 13 derived traits significantly associated with type 2 diabetes in meta-analysis (after Bonferroni correction). Adding IgG glycans to age and sex increased the AUC from 0.542 to 0.734. Adding them to the extensive model did not substantially improve the AUC. The AUC for IgG glycans alone was 0.729. Several IgG glycans and traits firmly associate with type 2 diabetes, reflecting a pro-inflammatory and biologically-aged state. IgG glycans showed limited improvement of AUCs. However, IgG glycans showed good prediction alone, indicating they may capture information of combined covariates. The associations found may yield insights in type 2 diabetes pathophysiology. This work shows that IgG glycomic changes have biomarker potential and may yield important insights into pathophysiology of complex public health diseases, illustrated here for the first time in type 2 diabetes.

The well-known backpropagation learning algorithm is implemented in a field-programmable gate array (FPGA) board and a microcontroller, focusing in obtaining efficient implementations in terms of a resource usage and computational speed. The algorithm was implemented in both cases using a training/validation/testing scheme in order to avoid overfitting problems. For the case of the FPGA implementation, a new neuron representation that reduces drastically the resource usage was introduced by combining the input and first hidden layer units in a single module. Further, a time-division multiplexing scheme was implemented for carrying out product computations taking advantage of the built-in digital signal processor cores. In both implementations, the floating-point data type representation normally used in a personal computer (PC) has been changed to a more efficient one based on a fixed-point scheme, reducing system memory variable usage and leading to an increase in computation speed. The results show that the modifications proposed produced a clear increase in computation speed in comparison with the standard PC-based implementation, demonstrating the usefulness of the intrinsic parallelism of FPGAs in neurocomputational tasks and the suitability of both implementations of the algorithm for its application to the real world problems.