Skövde Artificial Intelligence Lab
The Skövde Artificial Intelligence Lab (SAIL) conducts research within applied AI in close collaboration with industry. Current research activities are focused on reasoning under uncertainty, deep learning, visual analytics, transparent decision support, data privacy and recommender systems. In terms of education, SAIL is responsible for the University's master program in Data Science, the first of its kind in Sweden.
Reasoning under Uncertainty
SAIL has been conducting research regarding uncertainty representations, Bayesian theory, fuzzy sets and imprecise probability for a number of years. Within this context the group has explored how different uncertainty representation affect the end conclusions drawn from models/data and also formal aspects of decision making, including topics such as combination and aggregation of information. We have contributed in the development of the theory of fuzzy (non-additive) measures and integrals, and we have used them in different applications. This line of research links with measure theory. The group maintains an R-package where different uncertainty formalisms can be utilized.
Researchers: Nikolas Alexander Huhnstock, Alexander Karlsson, Vicenç Torra.
Deep learning is a group of techniques which all aim to find interesting useful patterns in complex data, e.g., high-dimensional big data. Examples of techniques used within the area are restricted Boltzmann machines, deep belief nets etc.. Recently, SAIL has started to approach this research field by exploring how different methods within this area can handle different data types and uncertainty as well as how the results from the methods can be interpreted.
Researchers: Denio Duarte, Nikolas Alexander Huhnstock, Göran Falkman, Alexander Karlsson, Gunnar Mathiason, Maria Riveiro, Niclas Ståhl.
Visual analytics is the science of analytical reasoning facilitated by interactive visual interfaces. It aims to aid the process of sense-making when dealing with complex data (in terms of size, dimensions and relations), by visually unveiling both its underlying features as well as the logic behind data mining algorithms. We conduct research in the areas of interactive data analysis, information visualization, visual analytics and human computer interaction, developing novel approaches to the analysis and visualization of large multidimensional data sets for decision support.
Researchers: Juhee Bae, Göran Falkman, Tove Helldin, Ulrika Ohlander, Maria Riveiro, Zhen Rong, Elio Ventocilla.
Transparent Decision Support
For a person it is often difficult to understand the internal reasoning processes of an automated decision support system. However, it is often the person who needs to make the final decision based on the system’s recommendations. Therefore, the system should make available the information necessary for a person to understand why a decision is recommended, what alternatives exist and how strongly these divagate from the most recommended decision. Transparent decision support is hence an integral part in our research. Recommended decisions are often based on probabilities, however, in a situation where there exist uncertainties regarding events, it is not always straightforward to select the most suitable decision. One option might be to include further context information, which often can be provided by a human expert. Hence human-machine interaction will benefit from an approach to transparent decision support where human and machine work together as a team.
Researchers: Göran Falkman, Tove Helldin, Alexander Karlsson, Gunnar Mathiason, Joe Steinhauer.
Data privacy (privacy preserving data mining, statistical disclosure control and privacy enhancing technologies) studies approaches to ensure confidentiality when data has to be published or transferred to third parties for their analysis. We have developed methods for data protection, as well as measures for disclosure risk and information loss (data utility). We have extensively used machine learning/data mining within the context of data privacy. In particular, clustering and classification algorithms have been used to measure information loss, and distance (metric) learning methods have been developed and applied to measure disclosure risk. We funded and are editors of the journal Transactions on Data Privacy.
Researchers: Per Gustavsson, Vicenç Torra.
At SAIL, research in recommender systems encompasses user modeling and personalization. Recommender systems tailor the experience of information access and delivery systems for their users by identifying the correct information for the right user at the right time. The information need of each user is specified by the user's context, the consumption device, and various other user-specific parameters. Recommender systems research at SAIL focuses on aspects related to replication, reproducibility and evaluation.
Researchers: Rakesh Rana, Alan Said.
SAIL is also responsible for Sweden's first Master program in Data Science.