Download Biological and artificial intelligence environments: 15th by Bruno Apolloni, Maria Marinaro, Roberto Tagliaferri PDF

By Bruno Apolloni, Maria Marinaro, Roberto Tagliaferri

The publication stories the complaints of the fifteenth Italian workshop on neural networks issued by way of the Italian Society on Neural Networks SIREN. The durability recipe of this convention stands in 3 details that generally renders the interpreting of those complaints so attention-grabbing as attractive. 1. the themes of the neural networks is taken into account an charm pole for a collection of researches headquartered at the inherent paradigm of the neural networks, instead of on a particular device completely. therefore, the subsymbolic administration of the knowledge info content material constitutes the major function of papers in a number of fields resembling trend popularity, Stochastic Optimization, studying, Granular Computing, etc, with a distinct bias towards bioinformatics operational functions. An excerpt of these types of issues could be present in the booklet. 2. although controlled at household point, the convention draws contributions from international researchers in addition, in order that within the publication the reader may possibly trap the flavour of the cutting-edge within the overseas neighborhood. three. The convention is a gathering of acquaintances to boot. hence the papers typically replicate a peaceful surroundings the place researchers meet to generously trade their notion and clarify their real ends up in view of a standard cultural turning out to be of the group.

Show description

Read or Download Biological and artificial intelligence environments: 15th Italian workshop on neural nets, WIRN VIETRI 2004 PDF

Best italian books

Extra info for Biological and artificial intelligence environments: 15th Italian workshop on neural nets, WIRN VIETRI 2004

Example text

30 expression data, together with the high sensitivity required for diagnostic problems, makes the classification of malignant and normal samples very challenging from a machine learning point of view. , 2002], that can be useful both to select the genes more related to malignancies and to enhance the discrimination power between normal and malignant tissues. , 2004] based on random subspace ensembles [Ho, 1998], that is sets of learning machines trained on randomly chosen subspaces of the original input space.

The final number of candidates obtained by both systems described above is still quite high, and needs a further reduction in order to be useful for clinical purposes. To perform this reduction we experimented neural networks whose input is the sub-image of the candidate itself created by extracting from the original radiograph an area of dimension 400 by 400 centered on the centroid of the candidate region considered. Several experiments have been done using as input down-sampled versions of the sub-images created.

0000 Sens. 8500 —— Spec. 7727 —— We considered different random subspaces of dimensionality from 2 to 2n−1 , randomly drawn from each 2n -dimensional gene space selected from the input space through the Golub’s method, while varying n between 5 and 10. According to Skurichina e Duin [Skurichina and Duin, 2002] we applied linear SVMs as base learners. Indeed they showed that random subspace ensembles are effective with linear base learners characterized by a decreasing learning curve (error) with respect to the cardinality n, especially when the dimensionality Feature selection combined with random subspace ensemble 33 is much larger than the cardinality.

Download PDF sample

Rated 4.89 of 5 – based on 34 votes