Ph.D.

Systematics and computer science

The project of expert systems in plant pathology was finished, and J. Le Renard joined M.N.H.N. as a computer scientist who wanted to apply his experience to Systematics for the Laboratory of Biology of Marine Invertebrates and Malacology (L.B.I.M.M.). I met again M. Manago who had made his Ph.D with Y. Kodratoff at LRI, university of Paris Orsay, and who was already involved in the INSTIL project. He had developed an ID3 like decision tree generator named KATE and wanted to link it with Case-Based Reasoning techniques. For doing my Ph.D., we decided to create a society named Acknowledge in 1991. I could thus postulate on a specific tripartite French convention called CIFRE between a laboratory, a society and a candidate. For the Systematics domain, we chose to work with Pr. C. Levi who is a renowned expert on marine sponges at M.N.H.N and distinguished Professor in life sciences.

On the computer science research side, I joined the Data Analysis team of INRIA Rocquencourt (CLOREC) and of University of Paris Dauphine (LISE-CEREMADE) with Pr. E. Diday. At this occasion, I had the chance to meet J. Lebbe and R. Vignes at INRIA, who worked also in research on computer aided Systematics. Going one day per week at INRIA Rocquencourt was nice for making a lot of discussions with them about classification and identification in biology, and search for computer science papers in the document center. At this moment, a book on Symbolic and Numeric data analysis emphasized the convergence between AI and data analysis techniques in France on the representational aspects of knowledge.

We wrote a paper in 1992 on the way to acquire descriptive knowledge for classification and identification tasks in Systematics, and another one to explain how to solve real world problems with KATE. Then, I participated with our renamed society AcknoSoft to the new European project that was following INSTIL, namely INRECA (INduction and REasoning from CAses). The industrial partners were from Irland (IMS) and Germany with TecInno bringing a CBR system, S3-CASE, a follow up of PATDEX developped by the university of Kaiserslautern, Artificial Intelligence - Knowledge-Based Systems Group. At the beginning of the project, the ideas of convergence between induction and CBR were applied to the induction and reasoning from cases on marine sponges. A case description language was conceived : CASUEL (Common Case representation Language) for dealing with object-oriented structured descriptions in the frame of Case-Based Reasoning.

But in 1993, the INRECA project turned out to be more oriented on industrial applications and integration of induction and CBR for the enhancement of Acknosoft (now named Kaïdara) and TecInno (now Empolis). This was indeed a more profitable choice than representing the complexity of biological objects for the enhancement of life sciences. Henceforth my contribution on knowledge acquisition of robust and structured cases with HyperQuest was becoming less important than focusing on knowledge processing with a lot of flat data, i.e. mixing different methods such as k-nearest neighbors data comparison and decision tree based identification for mining data. This subject was the theme of another Ph.D. student from E. Diday, E. Auriol who joined the AcknoSoft society in 1993, and wrote a paper and defended a thesis on the integration of both techniques. An evaluation of different CBR systems can be found here in the frame of the INRECA study made by K.D. Althoff.

On my side, I decided that computer science for Systematics research was my prior objective, and I leaved the society thanks to an opportunity given by P. Gigord and H. Ralambondrainy to become a Teaching Assistant at IREMIA, at the university of Reunion Island. I defended my thesis untitled "Improving the robustness of description, classification and identification support systems of biological objects" at the university of Paris IX Dauphine on the 24th of May 1994.