Convolutional neural networks (CNNs) have shown outstanding image classification performance, having been successfully applied in several real‐world applications. However, there are performance limitations encountered in CNNs and few works have pointed out these limitations across the literature. Therefore, this survey aims to provide a comprehensive explanation of both the importance and performance limitations of CNNs on image classification tasks. In this paper, we start with a brief CNN background and follow the main CNN architectures along with their contributions and benchmark datasets. We propose a classification of the CNNs performance limitations, which are grouped into four categories: labeled datasets, translation invariance, adversarial attacks, and spatial relationship. In addition, we describe some approaches that are currently being developed to overcome these performance limitations. Finally, we introduce specific ongoing research to address these performance limitations through capsule networks (CapsNets) and CNNs combined with CapsNets.

Human activity inference is not a simple process due to distinct ways of performing it. Our proposal presents the SCAN framework for activity inference. SCAN is divided into three modules: (1) artifact recognition, (2) activity inference, and (3) activity representation, integrating three important elements of Ambient Intelligence (AmI) (artifact-behavior modeling, event interpretation and context extraction). The framework extends the roaming beat (RB) concept by obtaining the representation using three kinds of technologies for activity inference. The RB is based on both analysis and recognition from artifact behavior for activity inference. A practical case is shown in a nursing home where a system affording 91.35% effectiveness was implemented in situ. Three examples are shown using RB representation for activity representation. Framework description, RB description and CALog system overcome distinct problems such as the feasibility to implement AmI systems, and to show the feasibility for accomplishing the challenges related to activity recognition based on artifact recognition. We discuss how the use of RBs might positively impact the problems faced by designers and developers for recovering information in an easier manner and thus they can develop tools focused on the user.

Recent advancements in cybersecurity threats and malware have brought into question the safety of modern software and computer systems. As a direct result of this, artificial intelligence-based solutions have been on the rise. The goal of this paper is to demonstrate the efficacy of memory-optimized machine learning solutions for the task of static analysis of software metadata. The study comprises an evaluation and comparison of the performance metrics of three popular machine learning solutions: artificial neural networks (ANN), support vector machines (SVMs), and gradient boosting machines (GBMs). The study provides insights into the effectiveness of memory-optimized machine learning solutions when detecting previously unseen malware. We found that ANNs shows the best performance with 93.44% accuracy classifying programs as either malware or legitimate even with extreme memory constraints.

The objective of this work is to select a Machine Learning Technique (MLT) to develop a fault diagnosis scheme for the power switching devices of the High Frequency link (HF link) in a grid-tied Photovoltaic (PV) system, without increasing the total number of sensors, and being capable to operate online. Artificial Neural Network (ANN), Support Vector Machine (SVM), Probabilistic Neural Network (PNN) and Naive Bayes (NB) algorithms are considered to solve the problem of fault classification. These four MLTs are compared using the specificity and sensitivity indexes. The inputs of the models are obtained from the mean value of the signals given by the Discrete Wavelet Transform (DWT) of the dc link voltage and the power extracted from the PV panels. Support vector machine algorithm is chosen as the most suitable classifier to diagnose single and simultaneous open circuit faults with lower computational effort. Simulation and real-time hardware-based experimental tests demonstrate that the MLTs are suitable and reliable to diagnose open circuit faults in a wide range of irradiance levels, ranging from 200 W/m2 to 1000W/m2, even under 6 % and 12 % measurement errors, without increasing the overall system cost.