Deep metric learning for open-set human action recognition in videos

Neural Computing and Applications - Human action recognition (HAR) is a topic widely studied in computer vision and pattern recognition. Despite the success of recent models for this issue, most of...     

A neuromorphic architecture for cortical multilayer integration of early visual tasks

The characteristics and performance of a hierarchical neural architecture, inspired by models of mammalian visual cortex, are considered. The visual pathway from sensory space to the intermediate (cortical) representation is structured in three layers, with intra and interlayer connections through feedforward and recurrent pathways. These interconnections provide a complex perceptual organization that integrates the specific functional tasks performed by each layer. This improves the capabilities of the architecture in feature extraction and segregation, further providing clues on the information content of the intermediate representation (primal sketch). Applications to preattentive vision tasks (edge and contour extractions, texture analysis and boundary completion, and defect detection) are presented with satisfactory results.     

Test of Inseparability Criteria for Squeezed Number States of the Radiation Field

We investigate the efficiency of inseparability criteria in detecting the entanglement properties of two-mode non-Gaussian states of the electromagnetic field. We focus our study on the relevant class of two-mode squeezed number states. These states combine the entangling capability of two-mode squeezers with the non-Gaussian character (nonclassicality) of number states. They allow for some exact analytical treatments, and include as a particular case the two-mode Gaussian squeezed vacuum. We show that the generalized PPT criterion recently proposed by Shchukin and Vogel, based on higher order statistical moments, is very efficient in detecting the entanglement for this class of non-Gaussian states.     

Oxidation of ethyl alcohol in trickle bed reactors: Analysis of the conversion rate

The interpretation of results of a chemical reaction carried out in a trickle bed reactor is not immediate when the key reactants are volatile; in fact some authors state that the overall conversion rate increases at very low liquid flow rates, i.e. when the catalyst is unevenly wetted. In these conditions, the reactants may penetrate inside the catalyst pellets directly through the “dry” zones; therefore the mass transfer resistance is diminished and the conversion rate can be increased. To investigate this phenomenon tests on the catalytic oxidation of ethyl alcohol dissolved in water were performed at different operating conditions in a trickle bed reactor. An interpretation of the results was tried by using two different models based on partially wetted particles. This approach considers both the cases in which “dry” zones are active or not for mass transfer and mass transfer rates are affected or not by the chemical reaction.     

DTT functionalization of Ag particles for conducting adhesives

We have studied the influence of the organic molecule DL-Dithiothreitol on the properties of isotropic conductive adhesives for application in microelectronics. The Ag fillers were prepared from commercial micro flakes or submicron particles from waste jewelry. The results show that coating of the Ag recycled fillers with thiol avoided the agglomeration of the particles improving the dispersion and the electrical and mechanical properties. By comparing the samples, the better properties were obtained from adhesives with recycled Ag particles relative to the ones formulated with commercial Ag flakes and improvement can be attributed to the absence of the lubricant layer and smaller size of the particles produced from recycled silver.     

Counselors network for intrusion detection

Intrusion detection systems (IDSs) are a fundamental component of defense solutions. In particular, IDSs aim to detect malicious activities on computer systems and networks by relying on data classification models built from a training dataset. However, classifiers' performance can vary for each attack pattern. A common technique to overcome this issue is to use ensemble methods, where multiple classifiers are employed and a final decision is taken combining their outputs. Despite the potential advantages of such an approach, its usefulness is limited in scenarios where (i) multiple expert classifiers present divergent results, (ii) all classifiers present poor results due to lack of representative features, or (iii) detectors have insufficient labeled signatures to train their classifiers for a specific attack pattern. In this work, we introduce the concept of a counselors network to deal with conflicts from different classifiers by exploiting the collaboration among IDSs that analyze multiple and heterogeneous data sources. Empirical results demonstrate the feasibility of the proposed architecture in improving the accuracy of the intrusion detection process.