Learning to cooperate in distributed partially observable environments with no communication abilities poses significant challenges for multi-agent deep reinforcement learning (MARL). This paper addresses key concerns in this domain, focusing on inferring state representations from individual agent observations and leveraging these representations to enhance agents' exploration and collaborative task execution policies. To this end, we propose a novel state modelling framework for cooperative MARL, where agents infer meaningful belief representations of the non-observable state, with respect to optimizing their own policies, while filtering redundant and less informative joint state information. Building upon this framework, we propose the MARL SMPE$^2$ algorithm. In SMPE$^2$, agents enhance their own policy's discriminative abilities under partial observability, explicitly by incorporating their beliefs into the policy network, and implicitly by adopting an adversarial type of exploration policies which encourages agents to discover novel, high-value states while improving the discriminative abilities of others. Experimentally, we show that SMPE$^2$ outperforms a plethora of state-of-the-art MARL algorithms in complex fully cooperative tasks from the MPE, LBF, and RWARE benchmarks.

Motivated by the global unrest related to the COVID-19 pandemic, this demo paper presents a system for acquisition of COVID-related data from different, public sources, and interlinking under a common semantic data model at a fine level of granularity. The integrated data set contains data from several European countries, which come in different schemata, formats, granularity, and data integration acts as a facilitator towards querying data from different sources, joint data analysis, and identifying correlations at varying geographical level. Moreover, our work shows how such an integrated data set can be exploited to answer complex questions for the pandemic, also in combination with other data sets via federated queries. © 2021 Copyright held by the owner/author(s).

An ever-increasing number of applications in critical domains, such as maritime and aviation, generate, collect, manage and process spatio-temporal data related to the mobility of entities. This wealth of data can be exploited for various purposes, towards improving the safety of operations, reducing economical costs, and increasing dependability: The major issue to achieve these objectives is increasing predictability of moving objects' trajectories and events. To achieve this purpose in a data-driven way we need to exploit in integrated manners data from a variety of disparate and heterogeneous data sources, both streaming and archival, regarding – among other – surveillance, weather, and contextual data. Motivated by this fact, in this paper, we propose a framework for semantic integration of big mobility data with other data sources that are necessary to data analytics tasks, providing a unified representation of such data. Notable features of our framework include the real-time generation of data synopses of moving entities' trajectories, the efficient and flexible transformation of data from heterogeneous and big data sources in RDF, and the spatio-temporal link discovery between spatio-temporal entities in diverse data sources. The design and implementation of our framework uses big data technologies (Apache Flink and Kafka), and our experimental evaluation demonstrates the efficiency and scalability of the proposed framework using large, real-life datasets.

This paper proposes and evaluates time and space efficient methods for matching entities in large data sets based on effectively pruning the candidate pairs to be matched, using edit distance as a string similarity metric. The paper proposes and compares three filtering methods that build on a basi c blocking technique to organize the target data set, facilitating efficient pruning of dissimilar pairs. The proposed filtering methods are compared in terms of runtime and memory usage: the first method clusters entities and exploits the triangle inequality using the string similarity metric, in conjunction to the substring matching filtering rule. The second method uses only the substring matching rule, while the third method uses the substring matching rule in conjunction to the character frequency matching filtering rule. Evaluation results show the pruning power of the different filtering methods used, also in comparison to the string matching functionality provided in LIMES and SILK, which are state of the art frameworks for large scale link discovery.

We present a big data framework for the prediction of streaming trajectory data, enriched from other data sources and exploiting mined patterns of trajectories, allowing accurate long-term predictions with low latency. To meet this goal, we follow a multi-step methodology. First, we efficiently compress surveillance data in an online fashion, by constructing trajectory synopses that are spatio-temporally linked with streaming and archival data from a variety of diverse and heterogeneous data sources. The enriched stream of trajectory synopses is stored in a distributed RDF store, supporting data exploration via SPARQL queries. The enriched stream of synopses along with the raw data is consumed by trajectory prediction algorithms that exploit mined patterns from the RDF store, namely medoids of (sub-) trajectory clusters, which prolong the horizon of useful predictions. The framework is extended with offline and online interactive visual analytics tool to facilitate real world analysis in the maritime and the aviation domains.

We present a stream reasoning system for monitoring vessel activity in large geographical areas. The system ingests a compressed vessel position stream, and performs online spatio-temporal link discovery to calculate proximity relations between vessels, and topological relations between vessel and static areas. Capitalizing on the discovered relations, a complex activity recognition engine, based on the Event Calculus, performs continuous pattern matching to detect various types of dangerous, suspicious and potentially illegal vessel activity. We evaluate the performance of the system by means of real datasets including kinematic messages from vessels, and demonstrate the effects of the highly efficient spatio-temporal link discovery on performance.

S.192-197

We introduce the concept of time mask, which is a type of temporal filter suitable for selection of multiple disjoint time intervals in which some query conditions fulfil. Such a filter can be applied to time-referenced objects, such as events and trajectories, for selecting those objects or segments of trajectories that fit in one of the selected time intervals. The selected subsets of objects or segments are dynamically summarized in various ways, and the summaries are represented visually on maps and/or other displays to enable exploration. The time mask filtering can be especially helpful in analysis of disparate data (e.g., event records, positions of moving objects, and time series of measurements), which may come from different sources. To detect relationships between such data, the analyst may set query conditions on the basis of one dataset and investigate the subsets of objects and values in the other datasets that co-occurred in time with these conditions. We describe the desired features of an interactive tool for time mask filtering and present a possible implementation of such a tool. By example of analysing two real world data collections related to aviation and maritime traffic, we show the way of using time masks in combination with other types of filters and demonstrate the utility of the time mask filtering. Keywords: Data visualization, Interactive visualization, Interaction technique

Symbolic event recognition systems have been successfully applied to a variety of application domains, extracting useful information in the form of events, allowing experts or other systems to monitor and respond when significant events are recognised. In a typical event recognition application, however, these systems often have to deal with a significant amount of uncertainty. In this article, we address the issue of uncertainty in logic-based event recognition by extending the Event Calculus with probabilistic reasoning. Markov logic networks are a natural candidate for our logic-based formalism. However, the temporal semantics of the Event Calculus introduce a number of challenges for the proposed model. We show how and under what assumptions we can overcome these problems. Additionally, we study how probabilistic modelling changes the behaviour of the formalism, affecting its key property—the inertia of fluents. Furthermore, we demonstrate the advantages of the probabilistic Event Calculus through examples and experiments in the domain of activity recognition, using a publicly available dataset for video surveillance.

Ontology modularization concerns about extracting ontology units from ontologies and partitioning large ontologies to possibly interdependent ontology units. Each unit specifies a specific context for performing ontology maintenance, evolution and reasoning tasks, which nevertheless has to be combined with chunks of tasks performed in other units. The modularization task is affected by assumptions concerning the mutual relations between the domains covered by distinct units, as well as by the expressiveness of the language used for specifying knowledge in units and for connecting distinct units. This paper presents a tool for partitioning SHIQ ontologies into units. These units can be combined using subjective class-to-class correspondences, as well as by inter-unit link properties that can be subjected to cardinality restrictions, existential and universal quantifiers, be hierarchically related and be transitive. While the modularization algorithm incorporated in this tool implements a specific partitioning method, the underlying representation framework provides a range of modularization possibilities, from units connected via class correspondences, to highly intertwined units, combined with class correspondences and inter-unit properties associated with restrictions.

This paper addresses the problem of synthesizing ontology alignment methods by maximizing the social welfare within a group of interacting agents: Specifically, each agent is responsible for computing mappings concerning a specific ontology element, using a specific alignment method. Each agent interacts with other agents with whom it shares constraints concerning the validity of the mappings it computes. Interacting agents form a bipartite factor graph, composed of variable and function nodes, representing alignment decisions and utilities, respectively. Agents need to reach an agreement to the mapping of the ontology elements consistently to the semantics of specifications with respect to their mapping preferences. Addressing the synthesis problem in such a way allows us to use an extension of the max-sum algorithm to generate near-to-optimal solutions to the alignment of ontologies through local decentralized message passing. We show the potential of such an approach by synthesizing a number of alignment methods, studying their performance in the OAEI benchmark series.

This paper presents a method along with a set of measures for evaluating learned ontologies against gold ontologies. The proposed method transforms the ontology concepts and their properties into a vector space representation to avoid the common string matching of concepts and properties at the lexical layer. The proposed evaluation measures exploit the vector space representation and calculate the similarity of the two ontologies (learned and gold) at the lexical and relational levels. Extensive evaluation experiments are provided, which show that these measures capture accurately the deviations from the gold ontology. The proposed method is tested using the Genia and the Lonely Planet gold ontologies, as well as the ontologies in the benchmark series of the Ontology Alignment Evaluation Initiative.

This paper proposes a method for the semi-automatic semantic annotation of WSDL specifications, given ontologies related to the domain of services. The proposed method uses a synthesis of mapping methods to map input/output messages' parameters to ontology classes. Exploiting validated results provided by humans, the method learns via the gradient boosting learning algorithm to combine the individual mapping methods towards improving its accuracy. The aim is to mitigate difficulties concerning mappings and address limitations of other approaches, even in challenging cases, so as to assist human annotators to perform their work. The paper presents experimental results of the proposed methods.

This paper proposes a method for learning ontologies given a corpus of text documents. The method identifies concepts in documents and organizes them into a subsumption hierarchy, without presupposing the existence of a seed ontology. The method unco

One of the most challenging problems in grid environments concerns the matchmaking between resource requests and offers. As it happens in the physical economy, grid economy must be supported by services that locate resources based not only on their characteristics, but also on market-related properties, offerspsila and requestspsila properties and constraints, as well as on declarative specifications of peerspsila (providers and consumers) features. Resource retrieval in the context of a grid economy extends the notion of resource matchmaking to the process of discovering those markets that trade resources through market orders. This paper describes an ontology that represents resource orders (offers and requests) in a market-oriented resource retrieval process, showing preliminary results of its utilization for the retrieval of traded resources.

Determining the size of an ontology that is automatically learned from texts is an open issue. In this paper, we study the similarity between ontology concepts at different levels of a taxonomy, quantifying in a natural manner the quality of the ontology attained. Our approach is integrated in a method for language-neutral learning of ontologies from texts, which relies on conditional independence tests over thematic topics that are discovered using LDA.

This paper proposes a method for the mapping of ontologies that, in a greater extent than other approaches, discovers and exploits sets of latent features for approximating the intended meaning of ontology elements. This is done by applying the reverse generative process of the Latent Dirichlet Allocation model. Similarity between element pairs is computed by means of the Kullback-Leibler divergence measure. Experimental results show the potential of the method.

This paper proposes a method for learning ontologies given a corpus of text documents. The method identifies concepts in documents and organizes them into a subsumption hierarchy, without presupposing the existence of a seed ontology. The method uncovers latent topics in terms of which document text is being generated. These topics form the concepts of the new ontology. This is done in a language neutral way, using probabilistic space reduction techniques over the original term space of the corpus. Given multiple sets of concepts (latent topics) being discovered, the proposed method constructs a subsumption hierarchy by performing conditional independence tests among pairs of latent topics, given a third one. The paper provides experimental results over the GENIA corpus from the domain of biomedicine.

For groups of agents to act collaboratively, they need to recognize the need for collaboration, decide on the method to be followed for achieving goal states, assign responsibilities to subgroups and individuals, and so on, until responsibilities that can be fulfilled by individuals are reached. Aiming to support collaborative activity of humans within organized settings, this paper introduces a set of constructs for specifying organizational structures and introduces an explicit representation of individual and collaborative responsibilities within organizations. We conjecture that group members create common awareness towards recognizing the need for collaboration by forming group acceptances. Acceptances are formed by means of shared practices and beliefs of individual agents. The paper introduces state recognition recipes that drive group members within organizations to form acceptances, and thoroughly explains the exploitation of these recipes in conjunction to state achievement recipes for achieving goal states and fulfilling responsibilities collaboratively.

This paper focuses on the knowledge representation framework utilized by an integrated wastewater treatment expert system for diagnosing operational problems of an activated sludge plant. The system deals with events that may occur in all the units of an activated sludge plant and exploits on‐line measurements, observations formed from data provided by laboratory analyses and empirical observations in an integrated manner. The system provides assistance to human experts to control the activated sludge process. It has been tested and evaluated in the pilot activated sludge plant of the Water and Air Quality Laboratory in the University of the Aegean.

The need for multilingual and user-tailored interfaces imposes new requirements upon the software industry: software applications must “speak” the language of users. Language engineering and knowledge engineering can assist the development of such interfaces. This paper presents a methodology for the creation of a language-independent knowledge base (KB), which can be used for the development of multilingual and user-tailored interfaces. This KB contains knowledge about the user interface components and functions and its creation is part of a software internationalisation process. The methodology aims at reducing the cost of setting up and managing this KB, by exploiting the benefits of controlled language use in technical writing. A case study for the dynamic generation of multilingual and user-tailored diagnostic messages is presented. Finally, the paper discusses related approaches in the area of multilinguality as well as in the area of software internationalisation and localisation, summarises the main results, and presents our plans for further exploitation of the methodology.