Hybrid fuzzy morphological/linear perceptrons based on extreme learning with application in classification

• Grant number: 17/10224-4
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): August 01, 2017
• Effective date (End): July 31, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Peter Sussner
• Grantee: Israel Campiotti
• Host Institution: Instituto de Matemática, Estatística e Computação Científica (IMECC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

Hybrid morphological/linear neural networks combine morphological with linear operators. We recently introduced a feedforward artificial neural network representing a hybrid fuzzy morphological/linear perceptron called fuzzy dilation/erosion/linear perceptron (F-DELP). Following Pessoa's and Maragos' ideas, an appropriate smoothing was applied to overcome the non-differentiability of the fuzzy dilation and erosion operators employed in the proposed F-DELP models. These models were trained using versions of the traditional backpropagation algorithm and its parameters were selected by means of cross-validation. The resulting F-DELP was applied to some well-known classification problems, achieving satisfactory results compared with some competitive classifiers such as FARC-HD, Theta-FAMs, and an SVM. Instead of cross-validation and backpropagation, more advanced techniques for selecting the number and the types of modules and for optimizing the weights could be used. In this project, we will focus on the use of extreme learning for determining the weights of the network. This way, the problem of selecting the number of modules can also be partially circumvented. According to Huang et al., extreme learning (EL) is computationally inexpensive compared to evolutionary optimization algorithms and classical neural network training algorithms and generally leads to a good generalization performance without requiring some form of regularization in order to avoid overfitting. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

SUSSNER, PETER; CAMPIOTTI, ISRAEL. Extreme learning machine for a new hybrid morphological/linear perceptron. NEURAL NETWORKS, v. 123, p. 288-298, MAR 2020. (18/13657-1, 17/10224-4)