T2 - META-DATA and META-LEARNING of DISTRIBUTED EXPERIMENTS

Abstract:  

In certain applications, experiments distributed in time or space deal with the same subject and bring their lighting.  Of each experiment, one can build a prediction, a knowledge, according to various techniques of  learning. The problem is to answer the following question:  how automatically  to treat the results obtained  by learning on distributed experiments?

 The meta-learning answers this question. It is a question of building a meta-knowledge, metadata.  It makes it possible to define a final prediction and to explain the variations observed on the predictions resulting from each learning.

 This tutorial presents a state of the art of the meta-learning of distributed and independent experiments.  The strategy of voting is thorough and an original strategy, the meta-analytical strategy, is proposed.  The talk is based on experiments.

 PLAN

 introduction

 automatic learning of experiments

 concepts of meta-learning of distributed experiments

 multi-strategy of voting of the meta-learning of distributed experiments

 meta-analytical strategy of the meta-learning of distributed experiments

 conclusion

 bibliography  and  text 100 pages

Brief authors CV:    

Claude DUSSART  doctor pharmacy 1997,  doctor computer science 2002 thesis “meta-learning of distributed experiments, meta-analytical strategic”,  researcher laboratory LASS umr 5823 CNRS, university Claude Bernard Lyon1,  project “meta-learning of distributed experiments”. 

Claude PETIT  doctor 1981 thesis “model global of enterprise”, doctor 1993 thesis “model-building with learning”, Diploma HDR 1996 “meta-models of applications”, researcher laboratory LASS umr 5823 CNRS, university Claude Bernard Lyon1, chief of project meta-learning of distributed experiments, chief of project e-learning master science computer, researcher director CESH.

 back to Tutorials home page