The high dimensional nature of the experimental research space is increasingly present in a variety of scientific areas, and characterizes at the moment most of biochemical experimentation. High dimensionality may pose serious problems in design the experiments because of the cost and feasibility of a large number of experiments, as requested by classical combinatorial design of experiments. In this work we propose a predictive approach based on neural network stochastic models where the design is derived in a sequential evolutionary way. From a set of neural networks trained on an initial random population of experiments, the best predictive net is chosen on a different validation set of experiments and adopted to predict the unknown space. The experiments which satisfy a predefined optimality criterion are then chosen and added to the initial population to form a second generation of experiments. The algorithm then proceeds through generations with a stopping rule on the convergence of the result. This approach, that allows to investigate large experimental spaces with a very small number of experiments, has been evaluated with a simulation study and shows a very good performance also in comparison with the genetic algorithm's approach.

An evolutionary predictive approach to design high dimensional experiments / D. De March; M. Forlin; D. Slanzi; I. Poli. - STAMPA. - (2010), pp. 81-88. [10.1142/9789814287456_0007]

An evolutionary predictive approach to design high dimensional experiments

DE MARCH, DAVIDE;
2010

Abstract

The high dimensional nature of the experimental research space is increasingly present in a variety of scientific areas, and characterizes at the moment most of biochemical experimentation. High dimensionality may pose serious problems in design the experiments because of the cost and feasibility of a large number of experiments, as requested by classical combinatorial design of experiments. In this work we propose a predictive approach based on neural network stochastic models where the design is derived in a sequential evolutionary way. From a set of neural networks trained on an initial random population of experiments, the best predictive net is chosen on a different validation set of experiments and adopted to predict the unknown space. The experiments which satisfy a predefined optimality criterion are then chosen and added to the initial population to form a second generation of experiments. The algorithm then proceeds through generations with a stopping rule on the convergence of the result. This approach, that allows to investigate large experimental spaces with a very small number of experiments, has been evaluated with a simulation study and shows a very good performance also in comparison with the genetic algorithm's approach.
2010
9789814287449
Artificial Life and Evolutionary Computation: Proceedings of WIVACE 2008
81
88
D. De March; M. Forlin; D. Slanzi; I. Poli
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/419252
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? 2
social impact