Autonomous learning of parameters in differential equations

Abstract : We propose EDEN+ 1 a fully automatic learner of parameters in dynamical systems that selects automatically the next experiment to do in the laboratory to improve its performance. EDEN+ improves upon EDEN, an experimental design algorithm proposed in the context of DREAM 6 and 7 challenges, with several new features: ability to take into account experiments with different costs, Monte-Carlo Tree Search parallelization, global optimization for parameter estimation. An illustration of the behaviour of EDEN+ is given on one of the DREAM7 challenging problems.