COMPUTATIONAL KNOWLEDGE DISCOVERY IN SCIENTIFIC APPLICATIONS

In the last decades, the framework of scientific work has significantly changed. The quality of measurement instrumentation has increased, enabling precise and fast collection of relevant data. Communication technologies had enabled availability and integration of many information sources. Existing human knowledge has increased enormously, and thanks to Internet technologies, is practically accessible worldwide. The consequence is that scientists dispose by a lot of easily accessible information and knowledge, and they are under constant pressure to produce novel results faster. The only solution is application of computer tools not only for data collection, calculation, or visualization, but also for automated modeling and hypotheses construction.

If we have a good hypothesis then the way to generally accepted theory or product is relatively fast and straightforward. However, getting a good hypothesis is not a trivial task. Hypotheses construction ability has always been recognized as the most distinguishing property of good scientists and the process of hypotheses construction has been recognized as the most complicated human intellectual activity. Today, enormous quantities of information are available. Gene activity measurements that produce more than 20.000 numerical data from a single sample, or millions of images automatically collected in a space mission, or thousands of chemical compounds synthesized every day, are examples of the existing information flood. Humans are unable even to read or look at all this information but they are supposed to extract valuable conclusions from it. On the other side, it is even harder to realize how diverse and profound existing human knowledge is in, for example medicine or physics. Scientists are supposed to work interdisciplinary by effectively combining knowledge from different domains. Acquiring and systematization of the available knowledge seems already an impossible mission. Scientists definitely need help.

But scientists are not the only one. Today, effective and intelligent handling of information and knowledge is already needed in different fields, varying from manufacturing to banking, from health care to transportation. Knowledge based society of the future is impossible without computers being able to practically realize complex tasks like knowledge representation, knowledge discovery and decision support. Even though practical implementation of these tasks today may seem as a science fiction, from our previous experience in data analyses, data modeling, and artificial intelligence applications we have learnt that they, regardless how intellectually complex they may seem, consist of some computational parts that may be automated and actually better executed by machines than by humans. A good example may be the well known fact that addition and multiplication of numbers have been, a century or two ago, recognized as intellectual operations uniquely attributed to humans, and today we have calculators and computers executing these operations much faster and more reliable than humans could ever do.

Our goal is to detect, realize, and apply procedures that may help in complex information and knowledge handling. Having that goal in mind, we will mainly concentrate on practical scientific applications. We expect that through interdisciplinary collaboration of computer scientists and scientists working on very specific scientific problems, we have the best chance to recognize the problems, to learn how to solve them, and finally, if possible, to generalize the solutions. The targets are already selected. They are data driven knowledge discovery processes aimed at construction of plausible scientific hypotheses or models. We start from applications in scientific domains but real aim is development of the technological basis for the knowledge society of the future.

->hrvatski

The program is supported by Croatian Ministry of Science Education and Sports
contract number: 0982560
starting date: July 2007

Participating projects

Intelligent image features extraction in knowledge discovery systems - University of Zagreb, Faculty of Electrical Engineering and Computing

Real life data measurement and characterization - Rudjer Boskovic Institute

Machine learning algorithms and their applications - Rudjer Boskovic Institute

Computational Intelligence Methods in Measurement Systems - Rudjer Boskovic Institute

Predictive models in health care - University of Zagreb, Medical School, Andrija Stampar School of Public Health

Documents

Scientific hypothesis
Collaboration plan
First program meeting - IRB 03/09/2007 presentation