Statistical significance analysis, based on hypothesis tests, is a common approach for comparing classifiers. However, many studies oversimplify this analysis by simply checking the condition p-value < 0.05, ignoring important concepts such as the effect size and the statistical power of the test. This problem is so worrying that the American Statistical Association has taken a strong stand on the subject, noting that although the p-value is a useful statistical measure, it has been abusively used and misinterpreted. This work highlights problems caused by the misuse of hypothesis tests and shows how the effect size and the power of the test can provide important information for better decision-making. To investigate these issues, we perform empirical studies with different classifiers and 50 datasets, using the Student’s t-test and the Wilcoxon test to compare classifiers. The results show that an isolated p-value analysis can lead to wrong conclusions and that the evaluation of the effect size and the power of the test contributes to a more principled decision-making.

Learning from human demonstration (LfD), among many speedup techniques for reinforcement learning (RL), has seen many successful applications. We consider one LfD technique called human–agent transfer (HAT), where a model of the human demonstrator’s decision function is induced via supervised learning and used as an initial bias for RL. Some recent work in LfD has investigated learning from observations only, that is, when only the demonstrator’s states (and not its actions) are available to the learner. Since the demonstrator’s actions are treated as labels for HAT, supervised learning becomes untenable in their absence. We adapt the idea of learning an inverse dynamics model from the data acquired by the learner’s interactions with the environment and deploy it to fill in the missing actions of the demonstrator. The resulting version of HAT—called state-only HAT (SoHAT)—is experimentally shown to preserve some advantages of HAT in benchmark domains with both discrete and continuous actions. This paper also establishes principled modifications of an existing baseline algorithm—called A3C—to create its HAT and SoHAT variants that are used in our experiments.

Openness is a challenging property that may characterize multi-agent systems (MAS). It refers to their ability to deal with entities leaving and joining agent society over time. This property makes the MAS behaviour complex and difficult to study and analyze, hence the need for a representative model allowing its understanding. In this context, many models were defined in the literature and we propose to classify them into three categories: structural models, functional models and interactional models. The existing models were proposed either for representing structural openness or for modelling functional or interactional ones independently. But, none of them was oriented to represent MAS openness in a global way while considering its three aspects at once. Besides, each one was defined in order to realize a specific objective and in a particular domain of application. In this paper, we propose an evolving KAGR graph. The latter provides a common understanding of openness and unifies its structural, functional and interactional aspects in a generic way. Our model is finally tested and validated on a multi-agent rescue simulator.

Designing hierarchical reinforcement learning algorithms that exhibit safe behaviour is not only vital for practical applications but also facilitates a better understanding of an agent’s decisions. We tackle this problem in the options framework (Sutton, Precup & Singh, 1999), a particular way to specify temporally abstract actions which allow an agent to use sub-policies with start and end conditions. We consider a behaviour as safe that avoids regions of state space with high uncertainty in the outcomes of actions. We propose an optimization objective that learns safe options by encouraging the agent to visit states with higher behavioural consistency. The proposed objective results in a trade-off between maximizing the standard expected return and minimizing the effect of model uncertainty in the return. We propose a policy gradient algorithm to optimize the constrained objective function. We examine the quantitative and qualitative behaviours of the proposed approach in a tabular grid world, continuous-state puddle world, and three games from the Arcade Learning Environment: Ms. Pacman, Amidar, and Q*Bert. Our approach achieves a reduction in the variance of return, boosts performance in environments with intrinsic variability in the reward structure, and compares favourably both with primitive actions and with risk-neutral options.

Argumentation and eXplainable Artificial Intelligence (XAI) are closely related, as in the recent years, Argumentation has been used for providing Explainability to AI. Argumentation can show step by step how an AI System reaches a decision; it can provide reasoning over uncertainty and can find solutions when conflicting information is faced. In this survey, we elaborate over the topics of Argumentation and XAI combined, by reviewing all the important methods and studies, as well as implementations that use Argumentation to provide Explainability in AI. More specifically, we show how Argumentation can enable Explainability for solving various types of problems in decision-making, justification of an opinion, and dialogues. Subsequently, we elaborate on how Argumentation can help in constructing explainable systems in various applications domains, such as in Medical Informatics, Law, the Semantic Web, Security, Robotics, and some general purpose systems. Finally, we present approaches that combine Machine Learning and Argumentation Theory, toward more interpretable predictive models.

We propose a fully distributed actor-critic architecture, named diffusion-distributed-actor-critic Diff-DAC, with application to multitask reinforcement learning (MRL). During the learning process, agents communicate their value and policy parameters to their neighbours, diffusing the information across a network of agents with no need for a central station. Each agent can only access data from its local task, but aims to learn a common policy that performs well for the whole set of tasks. The architecture is scalable, since the computational and communication cost per agent depends on the number of neighbours rather than the overall number of agents. We derive Diff-DAC from duality theory and provide novel insights into the actor-critic framework, showing that it is actually an instance of the dual-ascent method. We prove almost sure convergence of Diff-DAC to a common policy under general assumptions that hold even for deep neural network approximations. For more restrictive assumptions, we also prove that this common policy is a stationary point of an approximation of the original problem. Numerical results on multitask extensions of common continuous control benchmarks demonstrate that Diff-DAC stabilises learning and has a regularising effect that induces higher performance and better generalisation properties than previous architectures.

An ontology formalises a number of dependent and related concepts in a domain, encapsulated as a terminology. Manually defining such terminologies is a complex, time-consuming and error-prone task. Thus, there is great interest for strategies to learn terminologies automatically. However, most of the existing approaches induce a single concept definition at a time, disregarding dependencies that may exist among the concepts. As a consequence, terminologies that are difficult to interpret may be induced. Thus, systems capable of learning all concepts within a single task, respecting their dependency, are essential for reaching concise and readable ontologies. In this paper, we tackle this issue presenting three terminology learning strategies that aim at finding dependencies among concepts, before, during or after they have been defined. Experimental results show the advantages of regarding the dependencies among the concepts to achieve readable and concise terminologies, compared to a system that learns a single concept at a time. Moreover, the three strategies are compared and analysed towards discussing the strong and weak points of each one.

Automated planning is the field of Artificial Intelligence (AI) that focuses on identifying sequences of actions allowing to reach a goal state from a given initial state. The need of using such techniques in real-world applications has brought popular languages for expressing automated planning problems to provide direct support for continuous and discrete state variables, along with changes that can be either instantaneous or durative. PDDL+ (Planning Domain Definition Language +) models support the encoding of such representations, but the resulting planning problems are notoriously difficult for AI planners to cope with due to non-linear dependencies arising from the variables and infinite search spaces. This difficulty is exacerbated by the potentially huge fully ground representations used by modern planners in order to effectively explore the search space, which can make some problems impossible to tackle.

This paper investigates two grounding techniques for PDDL+ problems, both aimed at reducing the size of the full ground representation by reasoning over the lifted, more abstract problem structure. The first method extends the simple mechanism of invariant analysis to limit the groundings of operators upfront. The second method proposes to tackle the grounding process through a PDDL+ to classical planning abstraction; this allows us to leverage the amount of research done in the classical planning area. Our empirical analysis studies the effect of these novel approaches over both real-world hybrid applications and synthetic PDDL+ problems took from standard benchmarks of the planning community; our results reveal that not only the techniques improve the running time of previous grounding mechanisms but also let the planner extend the reach to problems that were not solvable before.

Currently, a large amount of spatial and spatiotemporal RDF data has been shared and exchanged on the Internet and various applications. Resource Description Framework (RDF) is widely accepted for representing and processing data in different (including spatiotemporal) application domains. The effective management of spatial and spatiotemporal RDF data are becoming more and more important. A lot of work has been done to study how to represent, query, store, and manage spatial and spatiotemporal RDF data. In order to grasp and learn the main ideas and research results of spatial and spatiotemporal RDF data, in this paper, we provide a comprehensive overview of RDF for spatial and spatiotemporal data management. We summarize spatial and spatiotemporal RDF data management from several essential aspects such as representation, querying, storage, performance assessment, datasets, and management tools. In addition, the direction of future research and some comparisons and analysis are also discussed in depth.

Radio-Frequency Identification (RFID) system technology is a key element for the realization of the Industry 4.0 vision, as it is vital for tasks such as entity tracking, identification and asset management. However, the plethora of RFID systems’ elements in combination with the wide range of factors that need to be taken under consideration along with the interrelations amongst them, make the problem of identification and design of the right RFID system, based on users’ needs particularly complex. The research outlined in this paper seeks to optimize this process by developing an integrating schema that will encapsulate this information in a form that is both human and machine processible. Human readability will allow a shared understanding of the RFID technology domain; machine readability, automated reasoning engines to perform logical deduction techniques returning implicit information. For this purpose, the novel RFID System Configuration Ontology (RFID SCO) is developed. Hence, non-RFID experts are enabled to identify the most suitable RFID system according to their needs and RFID experts to retrieve all the relevant information required for the efficient design of the corresponding RFID system. The RFID SCO is validated and tested successfully against real-world scenarios provided by domain experts.

Introduced as an interdisciplinary area that combines multi-agent systems, data mining and knowledge discovery, agent mining is currently in practice. To develop agent mining applications involves a combination of different approaches (model, architecture, technique and so on) from software agent and data mining (DM) areas. This paper presents an investigation of the approaches used in the agent mining systems by deeply analyzing 121 papers resulting from a systematic literature review. An ontology was defined to capitalize the knowledge collected from this study. The ontology is organized according to seven main facets: the problem addressed, the application domain, the agent-related and the mining-related elements, the models, processes and algorithms. This ontology is aimed at providing support to decisions about agent mining application design.

Patterns have been extensively used to extract hypernym relations from texts. The most popular patterns are Hearst’s patterns, formulated as regular expressions mainly based on lexical information. Experiences have reported good precision and low recall for such patterns. Thus, several approaches have been developed for improving recall. While these approaches perform better in terms of recall, it remains quite difficult to further increase recall without degrading precision. In this paper, we propose a novel 3-phase approach based on sequential pattern mining to improve pattern-based approaches in terms of both precision and recall by (i) using a rich pattern representation based on grammatical dependencies (ii) discovering new hypernym patterns, and (iii) extending hypernym patterns with anti-hypernym patterns to prune wrong extracted hypernym relations. The results obtained by performing experiments on three corpora confirm that using our approach, we are able to learn sequential patterns and combine them to outperform existing hypernym patterns in terms of precision and recall. The comparison to unsupervised distributional baselines for hypernym detection shows that, as expected, our approach yields much better performance. When compared to supervised distributional baselines for hypernym detection, our approach can be shown to be complementary and much less loosely coupled with training datasets and corpora.

This paper presents a model of contrastive explanation using structural casual models. The topic of causal explanation in artificial intelligence has gathered interest in recent years as researchers and practitioners aim to increase trust and understanding of intelligent decision-making. While different sub-fields of artificial intelligence have looked into this problem with a sub-field-specific view, there are few models that aim to capture explanation more generally. One general model is based on structural causal models. It defines an explanation as a fact that, if found to be true, would constitute an actual cause of a specific event. However, research in philosophy and social sciences shows that explanations are contrastive: that is, when people ask for an explanation of an event—the fact—they (sometimes implicitly) are asking for an explanation relative to some contrast case; that is, ‘Why P rather than Q?’. In this paper, we extend the structural causal model approach to define two complementary notions of contrastive explanation, and demonstrate them on two classical problems in artificial intelligence: classification and planning. We believe that this model can help researchers in subfields of artificial intelligence to better understand contrastive explanation.