This paper presents a methodology that permits to automate binary classification using the minimum possible number of attributes. In this methodology, the success of the binary prediction does not lie in the accuracy of an algorithm but in the evaluation metrics, which give information about the goodness of fit; which is an important factor when the data batch is unbalanced. The proposed methodology assesses the possible biases in identifying one algorithm as the best performer when considering the goodness of fit of an algorithm through evaluation metrics. The dimension of data has been reduced through the cumulative explained variance. Then, the performance of six machine learning classification models has been compared through Matthew correlation coefficient (MCC), area under curve – receiver operating characteristic (ROC-AUC), and area under curve – precision-recall (AUC-PR). The results show graphically and numerically how the evaluation metrics interfere with the most optimal outcome of an algorithm. The algorithms with the best performance in terms of evaluation metrics have been random forest and gradient boosting. In the imbalanced datasets, MCC has provided better prediction results than ROC-AUC or AUC-PR. The proposed methodology is adapted to the case of bankruptcy prediction.

Recently, many question answering systems that derive answers from linked data repositories have been developed. The purpose of this survey is to identify the common features and approaches of the semantic question answering (SQA) systems, although many different and prototype systems have been designed. The SQA systems use a formal query language like SPARQL and knowledge of a specific vocabulary. This paper analyses different frameworks, architectures, or systems that perform SQA and classifies SQA systems based on different criteria.

Artificial neural networks (ANNs) can be employed as controllers for robotic agents. Their structure is often complex, with many neurons and connections, especially when the robots have many sensors and actuators distributed across their bodies and/or when high expressive power is desirable. Pruning (removing neurons or connections) reduces the complexity of the ANN, thus increasing its energy efficiency, and has been reported to improve the generalization capability, in some cases. In addition, it is well-known that pruning in biological neural networks plays a fundamental role in the development of brains and their ability to learn. In this study, we consider the evolutionary optimization of neural controllers for the case study of Voxel-based soft robots, a kind of modular, bio-inspired soft robots, applying pruning during fitness evaluation. For a locomotion task, and for centralized as well as distributed controllers, we experimentally characterize the effect of different forms of pruning on after-pruning effectiveness, life-long effectiveness, adaptability to new terrains, and behavior. We find that incorporating some forms of pruning in neuroevolution leads to almost equally effective controllers as those evolved without pruning, with the benefit of higher robustness to pruning. We also observe occasional improvements in generalization ability.

Algorithmic bias arises in machine learning when models that may have reasonable overall accuracy are biased in favor of ‘good’ outcomes for one side of a sensitive category, for example gender or race. The bias will manifest as an underestimation of good outcomes for the under-represented minority. In a sense, we should not be surprised that a model might be biased when it has not been ‘asked’ not to be; reasonable accuracy can be achieved by ignoring the under-represented minority. A common strategy to address this issue is to include fairness as a component in the learning objective. In this paper, we consider including fairness as an additional criterion in model training and propose a multi-objective optimization strategy using Pareto Simulated Annealing that optimizes for both accuracy and underestimation bias. Our experiments show that this strategy can identify families of models with members representing different accuracy/fairness tradeoffs. We demonstrate the effectiveness of this strategy on two synthetic and two real-world datasets.

Day by day, new intelligent systems and autonomous machines are being developed to help and assist humans in a myriad of activities ranging from smart manufacturing to smart cities. Such new-generation intelligent systems need to work in teams and communicate with humans and other agents/robots to share information and coordinate activities. Furthermore, there is an increasing demand from government agencies and the private sector alike to use Unmanned Aerial Vehicles (UAVs), Unmanned Ground Vehicles (UGVs), Unmanned Surface Vehicles (USVs), and Autonomous Underwater Vehicles (AUVs) for tasks including search and rescue, surveillance, and monitoring. As these intelligent systems have to interact with humans in several scenarios involving multi-agent collaboration, data collection, and decision-making, it is urgent to discuss the technical as well as the ethical aspects in their design and function. Hence, ontology-based models for this Robotics and Automation (R&A) domain have the potential to enable a clear communication among the different intelligent systems and stakeholders, the formulations of standards, and the building of AI-based and robotics systems with full alignment with what stakeholders expect from these intelligent systems, in terms of economical benefits and enhanced human well-being.

Conventional reinforcement learning focuses on problems with single objective. However, many problems have multiple objectives or criteria that may be independent, related, or contradictory. In such cases, multi-objective reinforcement learning is used to propose a compromise among the solutions to balance the objectives. TOPSIS is a multi-criteria decision method that selects the alternative with minimum distance from the positive ideal solution and the maximum distance from the negative ideal solution, so it can be used effectively in the decision-making process to select the next action. In this research a single-policy algorithm called TOPSIS Q-Learning is provided with focus on its performance in online mode. Unlike all single-policy methods, in the first version of the algorithm, there is no need for the user to specify the weights of the objectives. The user’s preferences may not be completely definite, so all weight preferences are combined together as decision criteria and a solution is generated by considering all these preferences at once and user can model the uncertainty and weight changes of objectives around their specified preferences of objectives. If the user only wants to apply the algorithm for a specific set of weights the second version of the algorithm efficiently accomplishes that.

Traditional belief revision usually considers generic logic formulas, whilst in practical applications some formulas might even be inappropriate for beliefs. For instance, the formula $p \wedge q$ is syntactically consistent and is also an acceptable belief when there are no restrictions, but it might become unacceptable under restrictions in some context. If we assume that p represents ‘manufacturing product A’ and q represents ‘manufacturing product B’, an example of such a context would be the knowledge that there are not enough resources to manufacture them both and, hence, $p \wedge q$ would not be an acceptable belief. In this article, we propose a generic framework for belief revision under restrictions. We consider restrictions of either fixed or dynamic nature, and devise several postulates to characterize the behaviour of changing beliefs when new evidence emerges or the restriction changes. Moreover, we show that there is a representation theorem for each type of restriction. Finally, we discuss belief revision of qualitative spatio-temporal information under restrictions as an application of this new framework.

Reinforcement Learning (RL) methods often rely on gradient estimates to learn an optimal policy for control problems. These expensive computations result in long training times, a poor rate of convergence, and sample inefficiency when applied to real-world problems with a large state and action space. Evolutionary Computation (EC)-based techniques offer a gradient-free apparatus to train a deep neural network for RL problems. In this work, we leverage the benefits of EC and propose a novel variant of genetic algorithm called SP-GA which utilizes a species-inspired weight initialization strategy and trains a population of deep neural networks, each estimating the Q-function for the RL problem. Efficient encoding of a neural network that utilizes less memory is also proposed which provides an intuitive mechanism to apply Gaussian mutations and single-point crossover. The results on Atari 2600 games outline comparable performance with gradient-based algorithms like Deep Q-Network (DQN), Asynchronous Advantage Actor Critic (A3C), and gradient-free algorithms like Evolution Strategy (ES) and simple Genetic Algorithm (GA) while requiring far fewer hyperparameters to train. The algorithm also improved certain Key Performance Indicators (KPIs) when applied to a Remote Electrical Tilt (RET) optimization task in the telecommunication domain.

Knowledge-based systems and their ontologies evolve due to different reasons. Ontology evolution is the adaptation of an ontology and the propagation of these changes to dependent artifacts such as queries and other ontologies. Besides identifying basic/simple changes, it is imperative to identify complex changes between two versions of the same ontology to make this adaptation possible. There are many definitions of complex changes applied to ontologies in the literature. However, their specifications across works vary both in formalization and textual description. Some works also use different terminologies to refer to a change, while others use the same vocabulary to refer to distinct changes. Therefore, there is a lack of a unified list of complex changes. The main goals of this paper are: (i) present the primary documents that identify complex changes; (ii) provide critical analyses about the set of the complex changes proposed in the literature and the documents mentioning them; (iii) provide a unified list of complex changes mapping different sets of complex changes proposed by several authors; (iv) present a classification for those complex changes; and (v) describe some open directions of the area. The mappings between the complex changes provide a mechanism to relate and compare different proposals. The unified list is thus a reference for the complex changes published in the literature. It may assist the development of tools to identify changes between two versions of the same ontology and enable the adaptation of artifacts that depend on the evolved ontology.