GI-Forum WS 2013/14

Datum

Thema

Vortragender

Institution

The mission of Visual Analytics is to find ways to fundamentally improve the division of labour between humans and machines so that the computational power could amplify the human perceptual and cognitive capabilities. The term "Visual Analytics" stresses the key role of visual representations as the most effective means to convey information to human's mind and prompt human cognition and reasoning. Visual Analytics is defined as the science of analytical reasoning facilitated by interactive visual interfaces. It combines automated analysis techniques with interactive visualizations so that to support synergetic work of humans and computers.

In many areas of people's life and activities it is important to understand movement behaviours and mobility patterns of people, animals, vehicles, or other objects. Thanks to the recent advent of inexpensive positioning technologies, data about movement of various mobile objects or agents are collected in rapidly growing amounts. There is a pressing need in adequate methods for analysing these data and extracting relevant information.

Movement data are inherently complex as they involve (geographical) space and time. In addition to their own intrinsic complexities, these components are interdependent, which multiplies the overall complexity. As a result, movement data cannot be adequately modelled (at least at the present time) for a fully automatic analysis. At the same time, movement data, which are mostly acquired by automatic position tracking, are usually very poor semantically. The records basically consist of time stamps and coordinates. Semantic interpretations must emerge as a result of exploration and analysis where a human analyst plays the key role. Appropriate visual representations of movement data and outcomes from automated analysis procedures are paramount for this process.

The presentation gives an overview of how to analyse such data. For selected tasks, we propose scalable visual analytics methods. The work of the methods is illustrated using several examples of real-world datasets significantly differing in their properties. We analyse to what extent these and other existing methods cover the space of the types of movement data and the possible analysis tasks, identify the remaining gaps, and outline the directions for the future research.

12.11.2013
16 Uhr c.t.

In this talk I will discuss our on-going longitudinal ‘research-in-the-wild’ project with the rural village of Wray. The research has focussed on the user centered design of the Wray Photo Display – a situated display based system that enables village residents to upload images (typically photos) relating to their community for viewing by fellow residents and visitors to the village. Residents can also provide a response to pictures via the system’s commenting feature. A significant proportion of the images uploaded to the system relate to the cultural heritage of the village. For example, the first photo category to be added to the system by a member of the village was that of “Old Photos” and this category was shortly followed by the category “Wray Flood” (referring to the Wray Flood of 1967, a key event in the village’s history in which a number of buildings were destroyed). Note that individual residents of the village are responsible for moderating photo categories.

Our recently funded SHARC project (Investigating Technology Support for the Shared Curation of Local History in a Rural Community) is concerned with ways of usefully extending the currently deployed Photo Display system in order to support Locative Media experiences that can be produced and consumed by members of the village and visitors. Our initial steps are to explore scenarios to drive the technical advancement of the current system and to consider ways in which the current content repository (10 main photo categories with 26 subcategories, containing 2600 photo imahes and 450 comments) can be utilised in order to boot-strap the usefulness of new features. One example of this could be to provide category moderators with an easy-to-use tool that enables them to georeference images, (as well as carryout adding potentially useful other data and meta-data such as links to other photos, audio/textual descriptions of narratives/stories related to the picture/place, etc. etc.).

Speaker Bio
Keith Cheverst is a Reader in HCI at Lancaster University. His research over the last decade has focused on exploring the obdurate problems associated with the user-centred design of interactive systems (typically systems that utilise mobile and/or ubicomp technologies) that support communities in complex or semi-wild settings and the deployment and longitudinal study of these systems in order to gain insights into issues of user adoption and appropriation.

The talk will present ongoing work on the development of a spatial ontology for knowledge engineering.

An important result of the research was an evaluation of the degree of ambiguity in the meanings of natural language spatial vocabulary. Using the sentences of the text book as a corpus, the frequency of different meanings of spatial vocabulary was analysed. The insight gained from this analysis indicates that there may be a promising research area that combines the methodologies of logic-based knowledge representation and statistical corpus-based language analysis.

Another outcome of this work was the idea of developing a general purpose spatial ontology for knowledge engineering (SpOKE). Since biological information is extremely rich in spatial descriptions, the focus on biology had not limited the requirements of the spatial representation: if anything studying this domain leads one to consider spatial configurations that are more complex than one normally encounters in other domains. The idea of creating a spatial ontology is not new and various attempts have been made previously, including several for which very rigorous formal theories have been developed. However, the corpus based approach that was applied in the biology domain revealed some desiderata of a usable spatial ontology that are not supplied by logical rigour alone. These further requirements revolve around the relationship between the formal ontology representation and ordinary natural language. Of particular importance are two interconnected issues: (1) how do we reliably map from ordinary language spatial descriptions to a formal representation? (2) how do we clearly explain to knowledge engineers what the correct mapping should be? Much of the difficulty in solving these problems comes from the ambiguity and variability of language, which was highlighted by the corpus analysis.

In order to resolve these issues, the aim of the SpOKE ontology is to provide not just a formal representation, but a representation that can be explained clearly and applied consistently when encoding spatial information expressed in informal language.

Already the special version of Einstein's theory of relativity (SR) calls for a deep revision of our concept of space. If a space-wide instant of time exists at all, as I think it does, it must be relative to a frame of reference. This is why, at first sight, it is hard to see how one might reconcile SR with indeterminism. Indeterminism means: nature makes decisions by cutting off real alternatives. A few decades ago, indeterministically-minded modal logicians found out that a certain kind of necessity is positional when they were dealing with time: part of what wasn't fixed yesterday is fixed today. I propose a radical solution for reconciling SR with indeterminism: there is positional necessity in space, too. What is fixed here need not be fixed over there. This makes sense if we regard SR-space as a form of narration of nature's local decisions that helps us to coordinate them in order to tell nature's story. This idea can be provided a rigorous modeling in the framework of multi-dimensional modal logic.

A novel approach for spatial non-stationary interpolation is presented. This approach takes censored measurements, secondary information, and physically based neighbourhoods into account. The impact of the improvements of the geostatistical model are evaluated using regional groundwater quality parameters.

Spatial copulas are a multidimensional dependence model, that are capable of incorporating crisp and censored measurements. The dependence can deviate from Gaussian dependence and is independent of the marginal distribution. The model can be used for parameter estimation and interpolation purposes.

This presentation shows how all the three characteristics can be incorporated into a stochastic model: censored measurement, secondary information.?Censored measurements, such as measurements below some detection limit, are often ignored, but are incorporated via multidimensional conditional probabilities. Secondary information, such as land-use or hydrogeological units are incorporated. This secondary information has an influence on the distribution of concentration at each interpolation location. Physically-based delineated zones are taken into account for evaluating the influence of the neighbourhood on the measurement distribution. The pure distributions that are the basis for measurement concentration at each interpolation location depend on the composition of the neighbourhood. These pure distributions are estimated jointly using simulated annealing.

The improved quality of the interpolation is demonstrated by cross-validation results, also relating to the uncertainty of the interpolation.

The lecture discusses maps and mapping from the perspective of political geography and critical cartography. Following a Foucaultian line of argumentation maps and the tools and institutions which create them can be seen as parts of a powerful discourse producing spatial knowledge within society. Within this setting, the map can be regarded as a specific set of power-knowledge claim (Crampton and Krygier 2006, Glasze and Mose 2011).
The presentation gives an overview into some concepts in political geography (radical approaches, critical geopolitics) and critical cartography which might be useful to analyze and “deconstruct” the political geographies of maps and the political impact of mapping in geopolitical contexts. Additionally it presents some examples concerning the subtle visual power of maps and the “territorial trap” in such geopolitical representations.

Multiobjective optimization focuses on simultaneous optimization of multiple objective functions. As there exists no unique order in multiobjective space expert knowledge and preferences play a crucial role within the solution selection process. Different concepts for integrating preferences are discussed which basically can be incorporated prior or posterior to the optimization. Alternatively, a combination of both prior and posterior approaches can be applied to restrict the multiobjective space to the region which is of practical interest. Desirability Functions and Desirability Indices are presented as efficient tools for robust optimization of scalarized multiobjective optimization problems as well as for introducing bias within multiobjective evolutionary optimization algorithms. Finally, an outline of current research areas is given enhanced by directions for prospective research projects.