2 MSC, Universit Paris Diderot, UMR CNRS-7057, Paris 3 MAS, Centrale, - - PowerPoint PPT Presentation

▶

Oct 15, 2022 499 likes •566 views

Adel Mezine 1 , Artmis Llamosi 2 , Vronique Letort 3 , Michle Sebag 4 , Florence dAlch -Buc 1,5 1 IBISC, Universit dEvry - Val dEssonne, Evry 2 MSC, Universit Paris Diderot, UMR CNRS-7057, Paris 3 MAS, Centrale,

SLIDE 1

Adel Mezine1, Artémis Llamosi2, Véronique Letort3, Michèle Sebag4, Florence d’Alché-Buc1,5

1IBISC, Université d’Evry-Val d’Essonne, Evry 2MSC, Université Paris Diderot, UMR CNRS-7057, Paris 3MAS, Centrale, Châtenay-Malabry 4LRI UMR CNRS-8623, INRIA Université Paris-Sud, Université Paris Saclay 5LTCI UMR CNRS-5141, Telecom ParisTech, Université Paris Saclay

SLIDE 2

Motivations

Parameter and hidden state estimation Issues: 1. Some parameters are non-identifiable in practice

2. Perturbation experiments able to raise non-identifiability exist

but are very expensive.

2/5

SLIDE 3

EDEN: Experimental Design for Estimation in a Network

Autonomous learning Solution: Active learning that performs both system identification and sequential Design Of Experiments (DOE) under a fixed budget. Experimental Design: UCT-based active learning is applied to suggest the most promising sequence of experiments. (UCT = MCTS with UCB-policy) Parameter Estimation: The model is estimated from the current dataset using a global optimization procedure.

3/5

SLIDE 4

Dream7 Challenge

Perturbation Cost (credits) Knock-Out 800 Knock-Down 350 Over-Expression 450 Measurement Cost (credits) Mirco-array (50 points) 500 Mirco-array (100 points) 1000 Protein fluorescence (200 points) 400 Gel-shift assay (two parameters) 1600

Budget = 10,000 credits

Gene regulatory network Ordinary differential equations

4/5

SLIDE 5

Want to know more? Come see our poster!

5/5

SLIDE 6

References

Browne, C. B., Powley, E., Whitehouse, D., Lucas , S. M., Cowling, P, I., Rohlfshagen, P., . . .

& Colton, S.: A survey of Monte-Carlo tree search methods. Intelligence and AI (2012)

Kocsis, L., and Szepesvári, C.: Bandit based Monte-Carlo planning. ECML-06. (2006). Llamosi, A., Mezine, A., Sebag M., Letort V., d’Alché–Buc F. Experimental design in

dynamical system identification: a bandit-based active learning approach. ECML/PKDD Proceedings, LNCS, Springer (2014).

Meyer, P., et al. Network topology and parameter estimation: from experimental design

methods to gene regulatory network kinetics using a community based approach. BMC Systems Biology (2014).

Quach, M., Brunel, N., and d’Alché–Buc, F.: Estimating parameters and hidden variables

in non-linear state-space models based on odes for biological networks inference. Bioinformatics (2007).

Rolet, P., Sebag, M., and Teytaud, O.: Boosting active learning to optimality: A tractable

Monte-Carlo, billiard-based algorithm. ECML/PKDD (2009).