»Technology»Research
Our research activity
"For research and innovation, Bayesian networks have formalisms very interesting. The bayesian networks go further than other conventional systems."
The fields of application of bayesian networks are extremely broad. The software BayesiaLab and BEST provide access to bayesian networks technology to researchers and experts with their own research thematic, without being specialists of bayesian networks. Besides, these programs offer teachers very educational materials.
Bayesia works in collaboration with researchers worldwide. Bayesia studies their feedback to continually improve its products.
"I use BEST for training. This software is an excellent educational tool to teach students to make a diagnosis. "
"After explaining to students the trees and failure Markov chains, I find it easy to present graphs and dynamic models of BayesiaLab. They have a factored view of a problem which is very clear. And as can break the pattern, methodology is much easier to understand and easy to explain, even through extremely developed models."
Research projects
Project SKOOB
Object-oriented bayesian networks for modeling complex systems applications in risk management
Project funded by the ANR (Software Technologies) involving Bayesia, CHU Nancy, CRAN, EDF, ERPI, INERIS, LIP6 and Soredab
Risk is by nature closely related to hazard and, therefore, uncertainty. Bayesian networks are commonly known to be well suited for risk assessment due to their abilities to represent uncertain knowledge and to make rigorous probabilistic calculations. Partners of the SKOOB project design risk and safety analysis applications for various socio-economic systems of strategic importance (nuclear, food industries, medical or social organizations). Those applications imply simultaneous integration of various dimensions (technique, organisation, information, decision, finance) correlated with system's behaviour. This tends to increase models complexity, and leads to the development of increasingly larger Bayesian networks.
The SKOOB project focuses on both scientific and technical bottlenecks hindering the development of complex and complete system :
- Improvement of model authoring and engineering process, by facilitating complexity management, component reusability and collaborative work;
- Ability to design complex model even when both exact structure and borders (component counts, configurations, horizons of time) will be completely known only at run time ;
- Exploitation of complex model (for inference, machine learning, planning, etc) under limited resources (memory and time/computer power).
Possible answers to these requirements are :
- Object-oriented extension of the probabilistic modelling framework, by adding modularity and encapsulation to the traditional Bayesian network formalism ;
- Integration of various advanced aspects of object-oriented modelling such as inheritance, polymorphism and first order aspects;L'utilisation des aspects avancés de la modélisation orientée objets tels que l'héritage et le polymorphisme et/ou des éléments empruntés aux formalismes logiques de premier ordre ;
- Increase of algorithmic efficiency by taking advantage of the object-oriented characteristics of the models (pre-compilation of the network fragments, dynamic construction of the situation-specific Bayesian networks relevant to queries, exploitation of the parallelization potential) and by taking into account the available resources (any-time and any-space algorithms, approximations).
SKOOB project partners expect various outcomes :
- Risk management is one of the major challenges of our society. Success of SKOOB project will provide more powerful and reliable tools and methodologies to industrial partners (EDF, SOREDAB, University Hospital of Nancy), and to all other socio-economic actors confronted daily to risk management.
- This project will enforce a multidisciplinary community of risk management and Bayesian networks scientists (LIP6, CRAN, ERPI), contributing to the growth of this research field and its applications.
- This project will allow BAYESIA to enrich its software offer for risk analysis and decision support, and to achieve a substantial technological advantage over its competitors.