MSc on Information and Communication Systems Engineering





  • Tsoumakas Grigorios - Lecture: Learning from Multi-Label Data - Slides (PDF) - Related Links: Machine Learning and Knowledge Discovery (MLKD) group
    • Supervised learning has traditionally focused on the analysis of single-label data, where training examples are associated with a single label from a set of disjoint labels. However, training examples in several application domains, such as text and web mining, semantic annotation of images and video, music categorization into genres and emotions, functional genomics and directed marketing, are often associated with a set of labels. Such data are characterized as multi-label. This lecture will introduce the topic of learning from multi-label data, present the main existing learning methods, discuss interesting current research challenges, and demonstrate an open-source software library for using and further developing multi-label learning algorithms. The lecture will discuss issues such as learning to classify and rank from multi-label data, hierarchical multi-label classification, multi-label data sets and their statistics, multi-label evaluation metrics and multi-label dimensionality reduction.
  • Antonis Argyros - Lecture: Computational Vision: An introduction to problems, methods and applications
    • Computational Vision is a field of Artificial Intelligence aiming at the development of computational methods that will empower machines with visual perception capabilities. The design and the development of seeing machines has been a long sought interdisciplinary research goal with great theoretic and practical interest. The formulation and evaluation of computational vision methods may provide significant hints on the principles and mechanisms governing biological vision. Additionally, computational vision systems have numerous applications in almost every aspect of human life. In this lecture we will attempt a brief introduction to the goals, methods and state of the art results in the field of computational vision. Additionally, example application areas will be presented, with emphasis on vision for robotics and for smart environments. Finally, important, open research problems that constitute the current focus of the computational vision research community will be briefly discussed.
  • Kostas Karatzas - Lecture: Artificial Intelligence Applications for the Atmospheric Environment: recent developments and some research challenges
    • The investigation, analysis, understanding and management of environmental problems is one of the most fascinating domains of application for Artificial Intelligence. Among all environmental problems, that of air quality management is the one that is influencing the largest amount of citizens in Europe, as it is associate with the air that we all breathe. Air pollution (AP) is related to various substances, is affected by physical and chemical mechanisms of various spatial and temporal scales, and is regulated in terms of target values that are different to each other. Thus, AP requires for computational and knowledge management tools that are able to deal with its complex (and exiting from the scientific point of view) nature. Moreover, such methods should be able to deal with missing observation data, data of mixed nature (be it nominal, categorical, binary or other), and imitate the behavior and the "intelligence" of the phenomena that need to be modeled and simulated. This means that deterministic modeling, employing fluid mechanics, atmospheric chemistry and physics (the "traditional way for modeling AP") are not able to "catch" all the aspects of the problem. Other methods should be employed, that are able to deal with knowledge extraction and management, and are able to map knowledge into the "intelligence" of the algorithms that they apply. The lecture will address such methods, and will also provide input on recent developments and research challenges, from the policy making and environmental management field, as well as the field of emission reduction technologies and the automotive industry.
  • Ioannis Tsamardinos - Lecture: An introduction to Causal Discovery: A Bayesian Network Approach - Slides (PDF)
    • Discovery of causal knowledge is the ultimate goal of scientific research. Only when the causal mechanisms of a system have been uncovered can the technological means for manipulating the system be developed: drugs for treatment of disease in medicine, machines that compute in engineering or policies that reduce unemployment and inflation in economics. Observational data are typically significantly easier to obtain as computer technology becomes ubiquitous, and sensors and measuring technology is steadily improving. According to classical statistics, in cases where some of the data are observational, only correlations can be estimated. The induction of causal relations from observational data has traditionally been an anathema in statistics summarized in the well-known quote “correlation is not causation”. Nevertheless, theories and algorithms for causal induction from observational data, or mixed observational and experimental data under different sampling conditions exist. Apart from causal discovery per se, the theories and algorithms that originated from the field have yielded significant results with ramifications spanning a great portion of Machine Learning. These include the invention of Bayesian Networks employed for inference in Decision Support Systems and learning predictive or diagnostic models among others. The tutorial will first present the basic theory of causal discovery such as the Causal Markov Condition, the Faithfulness Condition, and the d-separation criterion, graphical models for representing causality such as Causal Bayesian Networks, Maximal Ancestral Graphs and Partial Ancestral Graphs. It will present prototypical and state-of-the-art algorithms such as the PC, FCI and ΗΙΤΟN for learning such models (global learning) or parts of such models (local learning) from data. The tutorial will also discuss the connections of causality to feature selection and present causal-based feature selection techniques. Finally, case-studies of applications of causal discovery algorithms will be presented.
  • Vassilios Verykios - Lecture: Frequent Itemset and Association Rule Hiding
    • Recent advances in data storage and collection devices have led the way to the upkeeping of vast amounts of data pertaining to different aspects of human life. This situation allows for effective techniques to be developed for processing and interrogating this data both in a scientific and in a commercial environment. The overabundance of such techniques creates suspicions to the data subjects regarding the appropriate use of the collected data, mostly because of the sensitive nature of these. To address this problem international and national authorities have enacted data protection laws which specify the rules governing the collection and handling of private information. In this lecture we will focus on privacy issues related to specific approaches of intelligent analysis of data, namely the data mining domain. We argue that it is not only the data that need to be protected from disclosure, but also the sensitive knowledge hidden in these data. Along these lines, we are going to present different approaches which are deemed necessary in order to protect the privacy of information during mining of the data. Our study will basically focus on privacy methodologies for frequent itemset and association rule mining, by building on the famous algorithm of Apriori proposed by Agrawal and Srikant. The approaches to be presented include a variety of hiding primitives, both heuristic and exact. Evaluation approaches of the presented methodologies will also be considered and discussed. Open issues and problems along with a roadmap for future research will also be provided.
  • Agisilaos Papantoniou - Lecture: Knowledge Desktop Environment – Organizational Memory exploitation by simulating Knowledge Intensive Tasks
    • The lecture addresses Knowledge Management in regards to its functional and technological aspects. It consists of the following topics
      1. Functional Level
      a. Definition of a Knowledge Intensive Task (KIT) – Presentation of KITs in the marine industry – Ship survey and repairs according to the Bureau Veritas specifications
      b. Presentation of a methodology for incorporating KIT models in a Knowledge Management Information system.
      2. Technological and implementation level
      a. Short theoretical background information on Organizational Memory Assistant Systems – correlation with formal Knowledge Based Systems
      b. Presentation of the multi-agent environment developed in the context of the European R&D project Knowledge Desktop Environment (KDE)
      i. Generic organizational model development methodology according to commonKADS guidelines
      ii. Modeling and Representation of Bureau Veritas Organizational Memory - Ontology development
      iii. Application of Problem Solving Methods
      iv. Axiom modeling through RDF(S) – Implementation of Inference mechanisms
      v. Architectural design and implementation of the multi-agent system – agent class definition and development
      vi. Graphical User Interface development – Interface Bubbling Mechanism
      vii. Results and operational scenarios
      3. Enrichment of KDE environment with modern Semantic Web technologies
      a. Short presentation of Rule Interchange Format (RIF) – Example: software agent communication language
      b. Short presentation of Web Ontology Language (OWL) – Enrichment of KDE ontology representation
  • Dimitris Vrakas - Lecture: An Introduction to Automated Planning: Languages, Algorithms and Applications
    • Automated Planning is the area of Artificial Intelligence that deals with problems, in which we are interested in finding a sequence of steps (actions) to apply to the world, in order to achieve a set of predefined objectives (goals), starting from a given initial state. In the past, planning has been successfully applied in numerous areas including robotics, space exploration, e-learning environments, semantic web, transportation logistics, marketing and finance, assembling parts, crisis management, e.t.c. The talk will begin with an introduction to the area of Automated Planning including its origins, some formal definitions of various concepts and a brief presentations of success stories in the application level. Then it will concentrate on knowledge representation issues including the STRIPS formalism and the PDDL language and solving methodologies including classical and neoclassical techniques. The talk will conclude with two state of the art applications in the areas of e-Learning and Semantic Web.
  • Kostas Stergiou - Lecture: Constraint satisfaction problems: State-of-the-art and the challenges ahead - Slides (PDF)
    • The constraint satisfaction problem (CSP) is a successful AI framework for representing and solving hard combinatorial problems from areas such as planning and scheduling, resource allocation, bioinformatics, configuration, etc. Research in the area of CSPs has lead to the developement of numerous highly sophisticated solvers that can nowdays handle large real problems. However, there are still many challenges that need to addressed in order to make this technology more widely understood and used. This lecture will first introduce the basic CSP concepts and solving methods focusing on the standard search and inference techniques employed by modern solvers. Then, a series of challenges and open problems that the CSP community faces will be presented.
  • Vassilios Peristeras, Lemonia Giantsiou - Lecture: The SemanticGov Portal
    • The process of discovering public services that can fulfill a citizen’s need can be a tedious and cumbersome task. Usually the citizen has a need, but does not know if and how has the Public Administration (PA) organized the coverage of this need and which PA services are currently available to address it. In order to overcome this problem, a semantically enabled portal is implemented, namely the SemanticGov Portal. The SemanticGov Portal has been implemented as a part of the overall SemanticGov platform, which is developed upon a Semantic Web Services (SWS) architecture, and plays the role of the Member State national e-Government portal. The Portal is loosely coupled with the rest of the SemanticGov platform. This practically means that the Portal acts as an entry point to the functionalities offered by the platform and is connected with it using Web Services. Therefore, the Portal could be connected with a different Web Service execution environment or even used without any execution environment as the functionalities that it provides are very important per se. In particular, the Portal provides three core functionalities: 1. Guides the citizens to find the exact public service version that fits to their personal characteristics/needs among the various service versions of a specific public service type. 2. Checks whether the citizen is eligible for a specific public service. 3. Provides complete and well-structured information for the identified service, i.e. inputs, outputs, provider, etc.
  • Nikolaos Mavridis - Lecture: Robots, Language, and Social Networks
    • Creating robots that can fluidly converse in natural language, and cooperate and sozialize with their human partners is a goal that has always captured human imagination. Furthermore, it is a goal that requires truly interdisciplinary research: engineering, computer science, as well as the cognitive sciences are crucial towards its realization. Challenges and current progress towards this goal will be illustrated through two real-world robot examples: the conversational robot “Ripley”, and the “FaceBots” social robots which utilize and publish social information on the FaceBook website. Finally, a quick glimpse towards novel educational and artistic avenues opened by such robots will be provided.