Turku Centre for Computer Science

About: Turku Centre for Computer Science is a facility organization based out in Turku, Finland. It is known for research contribution in the topics: Decidability & Word (group theory). The organization has 382 authors who have published 1027 publications receiving 19560 citations.

Information flow in intensive care narratives

Abstract: Fluent patient information flow is a prerequisite for clinical decision making. Our purpose is to identify unmet information needs in the flow of Finnish intensive care narratives in order to focus the development of natural language processing methods for this domain. Our data set consists of 516 authentic electronic patient records. First, we assess statistically the amount of narratives. We find that the amount is substantial: elective admission type and high nursing intensity contribute this. Second, we perform a content analysis. We observe that notes relevant for a given topic are scattered over the narratives, headings are inconsistent, and the flow from earlier narratives is fragmented. Consequently, support for gaining topical overviews is needed. Meeting this clinical need holds the promise of making narratives better accessible throughout a patient's stay and thereby improving clinical decision making and outcomes of care.

Implementing angelic nondeterminism

Abstract: We extend correctness and refinement reasoning methods in order to show how angelic nondeterminism can be systematically transformed into demonic nondeterminism or determinism. This kind of transformation is important because angelic nondeterminism assumes that an agent interested in establishing the postcondition will resolve the choices intelligently. When angelic nondeterminism is reduced into demonic nondeterminism or determinism, such intelligent choices are no longer necessary. Reducing angelic nondeterminism is generally not a refinement; however, when context is taken into consideration, it can be a correctness-preserving transformation.

Instruction Set Enhancements for High-Performance Multicore Execution on the REALJava Platform

Abstract: This paper presents instruction set enhancements for improving the performance of multicore execution of Java bytecode on the REALJava virtual machine. The instruction set is simplified by removing redundant bytecodes, and certain instruction sequences are transformed into so-called superinstructions in order to reduce the amount of interrupt requests produced by the REALJava co-processors and the amount of communication required between the host CPU and the co-processors. The technique greatly improves the performance of multi-threaded code and, to a lesser extent, single threaded code. The technique is applied as a preprocessing step during class loading and therefore has no negative impact on performance during actual execution.

Combining attack trees and fuzzy numbers in a multi-agent approach to fraud detection

Abstract: The fraud surveys carried out in the last five years by leading international consulting companies demonstrate that fraud is an increasing phenomenon depending most of all on behavioural aspects. In this paper, we introduce a multi-agent system called Fraud Interactive Decision Expert System (FIDES), which puts more emphasis on the evaluation of behavioural aspects of fraud detection according to the judgements expressed by experts (inspectors and auditors). FIDES combines think-maps, attack trees and fuzzy numbers under a Delphi-based team work support system and offers to the users a suitable way to better understand and manage fraud schemes.