Olga Zeydina and Bernard Beauzamy: Probabilistic Information Transfer

May 31, 2017 by

An Interview with Olga Zeydina and Bernard Beauzamy: Probabilistic Information Transfer

Michael F. Shaughnessy –

1) First of all, could you tell our readers a bit about yourselves, and your education and experience.

Olga Zeydina was a master student at Donetsk National University, Donetsk, Ukraine. Our Company, Société de Calcul Mathématique SA (Mathematical Modelling Company, Corp.) had a cooperation agreement with this University. Under this agreement, Olga came to France in 2006 to prepare a Ph.D, under the general title “Probabilistic Methods for Nuclear Safety”. This book is a consequence of her work.

Image result for Olga Zeydina and Bernard Beauzamy photos

Bernard Beauzamy was University Professor (1979-1995). He left his University position in 1995 in order to create SCMSA, and he has been chairman and CEO of this company since 1995.

2) What led up to you writing this book?

First, we collaborated on another book: Bernard Beauzamy et Olga Zeydina : Méthodes probabilistes pour la reconstruction de données manquantes, SCMSA, 2007 (Probabilistic methods for the reconstruction of missing data), which was well received. And all the contracts we treated at that time included the same preoccupation: there were missing data, and this is a constant preoccupation, in particular in the context of nuclear safety.

3) Now, what exactly is it all about?

The topic is the reconstruction of missing data. There are two situations: either you may rely upon other data, which are reliable, in order to complete your sample, or not. Our first book dealt with the first situation. The case when you have nothing outside your sample is of course much more delicate. You may of course complete using some assumptions, such as the phenomenon is linear, or periodical, or whatever you may say, but your conclusions will depend upon the specific model you introduce. So they may lead to a quarrel between experts.

4) Why is this topic and this book important?

Let us take an example, treated in detail in the book. There was some pollution in a harbor, and the French Company Total was suspected to be responsible for it. The data were of very poor quality: few samples, not well defined, not at regular places or depth. Some of these samples showed some pollution, some others showed nothing. So, there was a legal dispute. We were asked to come with our method for the reconstitution of the entire pollution in the harbor, because our method does not require any specific assumption. We came with our conclusion, which was unsatisfactory for both parties, and the dispute stopped.

Quite often, and especially in the context of Nuclear Safety, there is a need for robust methods, which do not rely upon specific assumptions. The Safety Authorities are becoming more and more reluctant about model assumptions: anyone may cheat.

 5) How much mathematical background is need to understand the content?

The “probabilistic information transfer” may be viewed as a tool, programmed in VBA (we call it EPH: Experimental Probabilistic Hypersurface”), and anyone may use this tool, without any specific mathematical knowledge. On the other hand, the complete understanding of the underlying theories (which rely upon the concept of entropy) is not so easy. But it should be understood that the more robust a theory is, the harder it is, mathematically speaking. You may declare that everything is linear, which requires very little mathematics, but this is not true in practice.

6) What have I neglected to ask?

Thanks for asking, anyway. We would be interested by an international collaboration upon such topics, and in particular validations of various reconstruction methods (krieging, and so on) upon various sets of data. We already have a number of case studies in which our EPH was compared to existing methods, but we need more. We consider that the validation of a tool, of a theory, may be done only at the international level.

Print Friendly, PDF & Email