Abstract
The traditional task of locating suspects using forensic sketches posted on public spaces, news, and social media can be a difficult task. Recent methods that use computer vision to improve this process present limitations, as they either do not use end-to-end networks for sketch recognition in police databases (which generally improve performance) or/and do not offer a photo-realistic representation of the sketch that could be used as alternative if the automatic matching process fails. This paper proposes a method that combines these two properties, using a conditional generative adversarial network (cGAN) and a pre-trained face recognition network that are jointly optimised as an end-to-end model. While the model can identify a short list of potential suspects in a given database, the cGAN offers an intermediate realistic face representation to support an alternative manual matching process. Evaluation on sketch-photo pairs from the CUFS, CUFSF and CelebA databases reveal the proposed method outperforms the state-of-the-art in most tasks, and that forcing an intermediate photo-realistic representation only results in a small performance decrease.

Abstract
SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of the most prominent indirect challenges advents from the mandatory use of face masks in a large number of countries. Face recognition methods struggle to perform identity verification with similar accuracy on masked and unmasked individuals. It has been shown that the performance of these methods drops considerably in the presence of face masks, especially if the reference image is unmasked. We propose FocusFace, a multi-task architecture that uses contrastive learning to be able to accurately perform masked face recognition. The proposed architecture is designed to be trained from scratch or to work on top of state-of-the-art face recognition methods without sacrificing the capabilities of a existing models in conventional face recognition tasks. We also explore different approaches to design the contrastive learning module. Results are presented in terms of masked-masked (MM) and unmasked-masked (U-M) face verification performance. For both settings, the results are on par with published methods, but for M-M specifically, the proposed method was able to outperform all the solutions that it was compared to. We further show that when using our method on top of already existing methods the training computational costs decrease significantly while retaining similar performances. The implementation and the trained models are available at GitHub.

Abstract
Background: Multiple sclerosis (MS) is a neurologic disease of the central nervous system which affects almost three million people worldwide. MS is characterized by a demyelination process that leads to brain lesions, allowing these affected areas to be visualized with magnetic resonance imaging (MRI). Deep learning techniques, especially computational algorithms based on convolutional neural networks (CNNs), have become a frequently used algorithm that performs feature self-learning and enables segmentation of structures in the image useful for quantitative analysis of MRIs, including quantitative analysis of MS. To obtain quantitative information about lesion volume, it is important to perform proper image preprocessing and accurate segmentation. Therefore, we propose a method for volumetric quantification of lesions on MRIs of MS patients using automatic segmentation of the brain and lesions by two CNNs. Methods: We used CNNs at two different moments: the first to perform brain extraction, and the second for lesion segmentation. This study includes four independent MRI datasets: one for training the brain segmentation models, two for training the lesion segmentation model, and one for testing. Results: The proposed brain detection architecture using binary cross-entropy as the loss function achieved a 0.9786 Dice coefficient, 0.9969 accuracy, 0.9851 precision, 0.9851 sensitivity, and 0.9985 specificity. In the second proposed framework for brain lesion segmentation, we obtained a 0.8893 Dice coefficient, 0.9996 accuracy, 0.9376 precision, 0.8609 sensitivity, and 0.9999 specificity. After quantifying the lesion volume of all patients from the test group using our proposed method, we obtained a mean value of 17,582 mm(3). Conclusions: We concluded that the proposed algorithm achieved accurate lesion detection and segmentation with reproducibility corresponding to state-of-the-art software tools and manual segmentation. We believe that this quantification method can add value to treatment monitoring and routine clinical evaluation of MS patients.

Abstract
It has been known for a while that the problem of multi-source domain adaptation can be regarded as a single source domain adaptation task where the source domain corresponds to a mixture of the original source domains. Nonetheless, how to adjust the mixture distribution weights remains an open question. Moreover, most existing work on this topic focuses only on minimizing the error on the source domains and achieving domain-invariant representations, which is insufficient to ensure low error on the target domain. In this work, we present a novel framework that addresses both problems and beats the current state of the art by using a mildly optimistic objective function and consistency regularization on the target samples.

Abstract
With the advent of self-driving cars, and big companies such as Waymo or Bosch pushing forward into fully driverless transportation services, the in-vehicle behaviour of passengers must be monitored to ensure safety and comfort. The use of audio-visual information is attractive by its spatio-temporal richness as well as non-invasive nature, but faces tile likely constraints posed by available hardware and energy consumption. Hence new strategies are required to improve the usage of these scarce resources. We propose the processing of audio and visual data in a cascade pipeline for in-vehicle action recognition. The data is processed by modality-specific sub-modules. with subsequent ones being used when a confident classification is not reached. Experiments show an interesting accuracy-acceleration trade-off when compared with a parallel pipeline with late fusion, presenting potential for industrial applications on embedded devices.

Abstract
The present work introduces a hybrid integration of the self-organizing map and the hidden Markov model (HMM) for anomaly detection in 802.11 wireless networks. The self-organizing hidden Markov model map (SOHMMM) deals with the spatial connections of HMMs, along with the inherent temporal dependencies of data sequences. In essence, an HMM is associated with each neuron of the SOHMMM lattice. In this paper, the SOHMMM algorithm is employed for anomaly detection in 802.11 wireless access point usage data. Furthermore, we extend the SOHMMM online gradient descent unsupervised learning algorithm for multivariate Gaussian emissions. The experimental analysis uses two types of data: synthetic data to investigate the accuracy and convergence of the SOHMMM algorithm and wireless simulation data to verify the significance and efficiency of the algorithm in anomaly detection. The sensitivity and specificity of the SOHMMM algorithm in anomaly detection are compared to two other approaches, namely HMM initialized with universal background model (HMM-UBM) and SOHMMM with zero neighborhood (Z-SOHMMM). The results from the wireless simulation experiments show that SOHMMM outperformed the aforementioned approaches in all the presented anomalous scenarios.