RESUMEN

In wake of COVID-19, the world has adapted to a new order. People have started wearing mask on their faces to prevent getting infected. The present face recognition models are no longer proving to be efficient in the current circumstances. This is because, most of the informative part of the face is covered by mask. The periocular recognition therefore holds the key to future of face recognition. However, the periocular region proves to be insufficiently enough to generate highly discriminative features. Also, most of the pre-COVID-19 algorithms fail to work in cases, where the number of training images available is very less. We propose a lightweight periocular recognition framework that uses thermo-visible features and ensemble subspace network classifier to improve upon the existing periocular recognition systems named as Masked Mobile Lightweight Thermo-visible Face Recognition (MmLwThV). The framework successfully improves the accuracy over a single visible modality by mitigating the effect of noise present in the thermo-visible features. The experiments on WHU-IIP dataset and an in-house collected dataset named, CVBL masked dataset, successfully validate the efficacy of our proposed framework. The MmLwFR framework is lightweight and can be easily deployed on mobile phones with a visible and an infrared camera.

RESUMEN

PURPOSE: The aims of this work are (1) to explore deep learning (DL) architectures, spectroscopic input types, and learning designs toward optimal quantification in MR spectroscopy of simulated pathological spectra; and (2) to demonstrate accuracy and precision of DL predictions in view of inherent bias toward the training distribution. METHODS: Simulated 1D spectra and 2D spectrograms that mimic an extensive range of pathological in vivo conditions are used to train and test 24 different DL architectures. Active learning through altered training and testing data distributions is probed to optimize quantification performance. Ensembles of networks are explored to improve DL robustness and reduce the variance of estimates. A set of scores compares performances of DL predictions and traditional model fitting (MF). RESULTS: Ensembles of heterogeneous networks that combine 1D frequency-domain and 2D time-frequency domain spectrograms as input perform best. Dataset augmentation with active learning can improve performance, but gains are limited. MF is more accurate, although DL appears to be more precise at low SNR. However, this overall improved precision originates from a strong bias for cases with high uncertainty toward the dataset the network has been trained with, tending toward its average value. CONCLUSION: MF mostly performs better compared to the faster DL approach. Potential intrinsic biases on training sets are dangerous in a clinical context that requires the algorithm to be unbiased to outliers (i.e., pathological data). Active learning and ensemble of networks are good strategies to improve prediction performances. However, data quality (sufficient SNR) has proven as a bottleneck for adequate unbiased performance-like in the case of MF.

Asunto(s)

RESUMEN

The power grid is going through significant changes with the introduction of renewable energy sources and the incorporation of smart grid technologies. These rapid advancements necessitate new models and analyses to keep up with the various emergent phenomena they induce. A major prerequisite of such work is the acquisition of well-constructed and accurate network datasets for the power grid infrastructure. In this paper, we propose a robust, scalable framework to synthesize power distribution networks that resemble their physical counterparts for a given region. We use openly available information about interdependent road and building infrastructures to construct the networks. In contrast to prior work based on network statistics, we incorporate engineering and economic constraints to create the networks. Additionally, we provide a framework to create ensembles of power distribution networks to generate multiple possible instances of the network for a given region. The comprehensive dataset consists of nodes with attributes, such as geocoordinates; type of node (residence, transformer, or substation); and edges with attributes, such as geometry, type of line (feeder lines, primary or secondary), and line parameters. For validation, we provide detailed comparisons of the generated networks with actual distribution networks. The generated datasets represent realistic test systems (as compared with standard test cases published by Institute of Electrical and Electronics Engineers (IEEE)) that can be used by network scientists to analyze complex events in power grids and to perform detailed sensitivity and statistical analyses over ensembles of networks.

Asunto(s)