Abstract
Saliency maps have been used as one possibility to interpret deep neural networks. This method estimates the relevance of each pixel in the image classification, with higher values representing pixels which contribute positively to classification.The goal of this study is to understand how the complexity of the network affects the interpretabilty of the saliency maps in classification tasks. To achieve that, we investigate how changes in the regularization affects the saliency maps produced, and their fidelity to the overall classification process of the network.The experimental setup consists in the calculation of the fidelity of five saliency map methods that were compare, applying them to models trained on the CIFAR-10 dataset, using different levels of weight decay on some or all the layers.Achieved results show that models with lower regularization are statistically (significance of 5%) more interpretable than the other models. Also, regularization applied only to the higher convolutional layers or fully-connected layers produce saliency maps with more fidelity. © 2020 IEEE.

Abstract

Abstract
Cyber-security is the practice of protecting computing systems and networks from digital attacks, which are a rising concern in the Information Age. With the growing pace at which new attacks are developed, conventional signature based attack detection methods are often not enough, and machine learning poses as a potential solution. Adversarial machine learning is a research area that examines both the generation and detection of adversarial examples, which are inputs specially crafted to deceive classifiers, and has been extensively studied specifically in the area of image recognition, where minor modifications are performed on images that cause a classifier to produce incorrect predictions. However, in other fields, such as intrusion and malware detection, the exploration of such methods is still growing. The aim of this survey is to explore works that apply adversarial machine learning concepts to intrusion and malware detection scenarios. We concluded that a wide variety of attacks were tested and proven effective in malware and intrusion detection, although their practicality was not tested in intrusion scenarios. Adversarial defenses were substantially less explored, although their effectiveness was also proven at resisting adversarial attacks. We also concluded that, contrarily to malware scenarios, the variety of datasets in intrusion scenarios is still very small, with the most used dataset being greatly outdated.

Abstract
e13070 Background: Bone is one of the main sites of breast cancer metastasis. Staging of this kind of disease spread can be performed in locally advanced cases with PET-CT in conjunction with Bone Scintigraphy. The purpose of this work is to compare the efficiency of bone metastasis detection between PET-CT and bone scintigraphy. Methods: Prospective analysis of locally advanced breast cancer patients treated in a Comprehensive Cancer Center between 2014 and 2019 that performed PET-CT and Bone Scintigraphy in the staging. Interval between the two exams could not exceed 2 months. Clinical and pathological characteristics of the disease were collected from electronic files and independently clinical images reports were considered to evaluate the ability of each imaging modalities to identify bone disease. In discrepancy cases a re-analysis of the images by two independent nuclear physicians was performed to validate the findings. Results: We analyzed 204 cases. The majority of them had ductal carcinomas (72.5%), cT2/3 (70%), cN1/2(61.8%) and G2/3 (94.6%), luminal B- like, HER2 positive disease (49.2%). In this cohort, bone metastasis was documented in 52 (25.5%) patients. PET-CT presented 97.0% of accuracy, surpassing the 94.1% presented by Bone Scintigraphy. The latter failed to correctly detect bone metastasis in 11 (5.4%) patients and only outperformed PET-CT in 3 (1.5%) patients. The main difference between the two modalities was the non-detection of cranium lesions in PET-CT images. Conclusions: PET-CT showed higher efficiency in bone metastasis detection than Bone Scintigraphy, probably because it detects lytic lesions. The non-detection of cranium ones can be harmful and so modifications in the image acquisition are required to improve the quality of PET-CT, avoiding other exams in bone staging.

Abstract
This paper proposes the linear frequency cepstral coefficients as highly discriminative features for anomaly detection in ball bearings using vibration sensor data. These features are based on cepstral analysis and are capable of encoding the patterns of a spectral magnitude profile. Incipient damages on bearings can grow rapidly under normal use resulting in vibration and harsh noise. If left undetected, this damage will worsen, leading to high maintenance costs or even injury. Multiple interferences in an industrial environment contaminate the signal, making it a challenge to correctly identify the bearings' condition. Many studies have attempted to overcome this issue at the signal level. However, the discriminative capacity of the current vibration signal features is still vulnerable to interference, which motivates this work. In order to demonstrate the benefits of these features, we (1) show that they are computationally efficient and suitable for real-time incremental training; (2) conduct discriminative analysis by evaluating the separability performance and comparing it with the state of the art; and (3) test the robustness of the proposed features under noise interference, which is ideal for use in the harsh operating conditions of industrial machinery. The data was obtained from a laboratory workbench setting that reproduces bearing fault scenarios. Results show that the proposed features are fast, competitive when compared to state-of-the-art features, and resilient to high levels of interference. Despite the higher performance when using the quadratic model, the proposed features remain highly discriminative when used with several other discriminant function.

Abstract
The dynamic behavior of urban mobility patterns makes matching taxi supply with demand as one of the biggest challenges in this industry. Recently, the increasing availability of massive broadcast GPS data has encouraged the exploration of this issue under different perspectives. One possible solution is to build a data-driven real-time taxi-dispatching recommender system. However, existing systems are based on strong assumptions such as stationary demand distributions and finite training sets, which make them inadequate for modeling the dynamic nature of the network. In this paper, we propose BRIGHT: a drift-aware supervised learning framework which aims to provide accurate predictions for short-term horizon taxi demand quantities through a creative ensemble of time series analysis methods that handle distinct types of concept drift. A large experimental set-up which includes three real-world transportation networks and a synthetic test-bed with artificially inserted concept drifts, was employed to illustrate the advantages of BRIGHT when compared to S.o.A methods for this problem.