Invited speakers

Paulo J.G. Lisboa

Unsupervised ML in Sports Analytics

Advances in data capture and cheap computer processing have made the analysis of data ubiquitous in sport and potentially of value for gaining a competitive advantage. Simple visualization methods readily show player segmentation according to position, from annotated match data. Yet, in-match measurements are made at increasingly fast sampling rates – how much information can be reliably derived from this data avalanche? The talk will focus on multivariate analysis (team formation pattern analysis, conditional independence maps) to generate actionable insights of potential value to football professionals. A more complex but still preliminary study involves segmenting periods of possession according to playing style. Even though this study deals with a fundamental aspect of play, it highlights the need for a more systematic taxonomy of playing styles, which builds on clustering and requires both knowledge discovery from data and interpretation by panels of experts.

Paulo Lisboa is Full Professor and Head of Department of Applied Mathematics at Liverpool John Moores University (UK). His research focus is advanced data analysis for decision support, in particular with applications to personalised medicine, public health, sports analytics and digital marketing. He chairs the Horizon2020 Advisory Group for Societal Challenge 1: Health, Demographic Change and Wellbeing, providing scientific advice to one of the world’s largest coordinated research programmes in health.

Tobias Schreck

Visual Analytics Techniques for Data Exploration: Visual Cluster Analysis, Interactive Data Modeling, and User Guidance

Visual Analytics aims to support users in interactive data exploration and pattern discovery. Visual Analytics methods rely on interactive data visualization, tightly coupled with automatic data analysis algorithms. After an introduction to foundations of Visual Analytics, we will discuss three examples of Visual Analytics approaches from our work. First, we will consider applications of the Self-Organizing Map algorithm for visual exploration of time series and trajectory data, including techniques for interactive map training. Then, we will consider approaches for visual retrieval and modeling of data patterns, relying on sketch-based search and interactive regression analysis. As a third line of research, we will discuss techniques for user guidance through the data exploration process, based on relevance feedback and indirect user input modalities. We will conclude by highlighting opportunities for future work in Visual Analytics.

Tobias Schreck is a Full Professor with the Institute for Computer Graphics and Knowledge Visualization at the Faculty for Computer Science and Biomedical Engineering of TU Graz, Austria. His main research interests are in 3D Object Retrieval and Processing, Visual Analytics, and Visual Search for Digital Library applications. Further application areas include large data exploration, visual search interfaces, 3D object search, and shape restoration in 3D Archeology.

Aïda Valls

Ontology-based Clustering for Managing Semantic Data and its Use in Tourism Applications

Datasets with multi-valued linguistic variables are becoming popular due to the facility in collecting information from users or objects by means of automatic text-based tools. The values of this kind of variables are words with a meaning that should be properly interpreted when applying unsupervised techniques, such as clustering. In this talk, I will present how to extend some clustering methods to be able to work with ontologies, which are data knowledge structures that provide semantic information about the concepts of a certain domain. Some applications in the field of Tourism will be shown, where the semantic interpretation of the data provides relevant information in order to build clusters that properly model the user’s needs. The use of ontologies permits to discover concept-based relations on the objects, discovering clusters with a semantic interpretation useful for tourism management.

Aïda Valls is an associate professor at the Department of Computer Science and Mathematics in Universitat Rovira i Virgili (URV), Tarragona, Spain. Her research interests include multiple criteria decision making, recommender systems, data mining and machine learning. Her work is mainly focused on the treatment of linguistic and semantic information. She has participated in several Spanish and EU research projects, with applications in Tourism, Environment Risk Management and Health Care.

Alessandro Sperduti

Self-organizing Maps for Structured Data: from Unsupervised to Deep Supervised Models

We start by introducing structured domains and the main computational issues associated with them. We then move to the introduction of a

general framework for the definition of the earlier deterministic SOMs models for dealing with structured data. We also show how the same ideas

can be applied in a probabilistic setting. All these models are unsupervised. However, SOMs can also be used as components in supervised models. Specifically, we show at least a couple of ways to use them to define kernels for structured data. Both deterministic and probabilistic approaches are discussed. Finally, we present more recent work on the application of SOMs as a component of Deep Convolutional Neural Networks for Graphs. Future directions of research are finally outlined.

Alessandro Sperduti is Full Professor at the Dipartimento di Matematica Pura ed Applicata, Universita degli Studi di Padova, Italy.