Extreme learning machine for a new hybrid morphological/linear perceptron
Introduction
Along with morphological associative memories (MAMs), morphological perceptrons (MPs) were among the first morphological neural network (MNN) models that appeared in the literature (Ritter et al., 1999, Ritter and Sussner, 1996, Ritter and Sussner, 1997, Ritter et al., 1998, Sussner, 1998, Sussner and Esmi, 2011, Sussner and Valle, 2006). The commonality shared by all MNNs is that their nodes, called “morphological neurons”, apply a “morphological operator” before applying an activation function. Unfortunately, there is no precise definition of a morphological operator as far as applications in image processing and analysis are concerned. According to Henk Heijman’s seminal book entitled “Morphological Image Operators” (Heijmans, 1994): “Any attempt to find a formal definition of a morphological operator, however, would lead inevitably to the following dilemma: either it would be too restrictive, excluding operators that should not be excluded a priori, or it would be too general, leading to a ‘theory of everything’”.
However, when resorting to the algebraic framework of mathematical morphology (MM), one finds exact definitions of the four types of elementary operators of MM on complete lattices. These four operators are erosion, dilation, anti-erosion, and anti-dilation (Banon & Barrera, 1993). Banon and Barrera have shown that every mapping between complete lattices can be expressed both as a supremum of pair-wise infima of erosions and anti-dilations and as an infimum of pair-wise suprema of dilations and anti-erosions. In view of this fact, it is interesting to observe that the overall computations at every node in the output layer of an MP can be represented as a supremum of pair-wise infima of gray-scale erosions and anti-dilations (Sussner & Esmi, 2011). The level curve of an infimum of a gray-scale erosion and an anti-dilation represents a hyperbox. Several constructive (hyperbox) learning algorithms for training MPs (Sussner, 1998), in particular competitive training algorithms (Sussner and Esmi, 2009a, Sussner and Esmi, 2009b, Sussner and Esmi, 2011), have appeared in the literature. Other hyperbox learning algorithms were used by Ritter, Urcid, Sossa and Guevara to train parameters of dendritic MNNs (Ritter and Urcid, 2003, Sossa and Guevara, 2014). In classification problems, each resulting hyperbox encloses patterns, all or most of which belong to a single class. Note that hyperbox-based algorithms were also independently developed by Kaburlasos et al. for training fuzzy lattice neurocomputing models (Kaburlasos et al., 2007, Kaburlasos and Petridis, 2000).
One of the main reasons why researchers devised constructive training algorithms in order to train morphological perceptrons or dendritic MNNs, instead of resorting to standard methods of non-linear optimization, is the fact that the morphological operations used in these models are not differentiable. In their recent approaches towards training dendritic morphological and hybrid morphological/linear neural networks via stochastic gradient descent (Hernández et al., 2018, Zamora and Sossa, 2017), Zamora, Sossa, et al. apparently dealt with locations where the partial derivatives do not exist set by setting the search directions equal to . In any case, the vector of search directions is not continuous, which is certainly a disadvantage. Also note that at locations where the function described by a dendrite is differentiable, all but one of its partial derivatives with respect to a weight are equal to .
An alternative for training the weights of morphological neurons without encountering these problems consists in smoothing the morphological operations. This approach was employed by Pessoa and Maragos in their hybrid morphological/rank/linear neural network (MRL-NN) (Pessoa & Maragos, 2000) as well as in Araújo et al.’s morphological and hybrid morphological/linear neural networks (Araújo, 2012, Araújo et al., 2017, Araújo and Sussner, 2010). Finally, note that MRL-NNs, other hybrid models, and dendritic MNNs can also be trained by using methods of evolutionary computation (Araújo and Ferreira, 2013, Arce et al., 2018).
Our approach for training hybrid morphological/linear neural networks differs greatly from all aforementioned approaches since it is based on Huang et al.’s extreme learning machine. According to Huang, Zhu, and Siew (2006), training a network using extreme learning machine (ELM) (Huang et al., 2006) is computationally inexpensive compared to evolutionary optimization (Simon, 2013) 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” (Bishop, 1995, Bishop, 2006, Ripley, 1996). The hybrid morphological/linear perceptron (HMLP) introduced in this paper is a feedforward neural network model that includes hidden morphological units taken from the previous MP models. The outputs of these morphological units and traditional semi-linear neurons are then combined using linear aggregation functions in the output neurons. Therefore, an ELM-based approach can effectively be employed in order to learn the weights of these linear combinations. In this paper, we apply our HMLP approach to a number of classification problems from the literature and compare the resulting classification rates with the ones of several related models.
The paper is organized as follows: Section 2 reviews some relevant concepts of lattice theory and mathematical morphology,including a few pertinent comments on gray-scale mathematical morphology. After discussing morphological perceptrons and their relationship with some related models from the literature, as well as providing some motivation, we introduce a new hybrid morphological/linear perceptron (HMLP) in Section 3. Then we present an extreme learning machine approach for HMLP training in Section 4 and compare the classification performance achieved by our model with related models from the recent literature in Section 5. We finish the paper with some concluding remarks.
Section snippets
Some relevant concepts of lattice theory and mathematical morphology
Lattice-theoretical concepts play an important role in mathematical morphology ever since complete lattices were established as an appropriate mathematical framework for MM (Heijmans, 1994, Heijmans and Ronse, 1990, Ronse, 1990, Serra, 1988). Therefore MNNs can be viewed as a lattice computing approach towards computational intelligence (Kaburlasos & Kehagias, 2014). Let us proceed by reviewing some basic concepts.
A partially ordered set or poset is a non-empty set together with a
From morphological to hybrid morphological/linear perceptrons: Some background and motivation
Note that the previous section refers to computations in since this set, together with the product partial order and addition, represents a complete lattice-ordered group extension (Sussner & Esmi, 2011), which is a suitable mathematical structure in order to establish links with mathematical morphology and minimax algebra (Cuninghame-Green, 1979, Heijmans, 1994). This said, both the weight vectors and the input patterns of morphological and hybrid morphological/linear neural networks have
Training hybrid morphological/linear perceptrons using extreme learning machine
The technical term “extreme learning machine”, coined by Huang et al. (2006), essentially refers to a class of feedforward computational models whose first layer of weights can be randomly selected and whose second layer of weights can be determined as the global minimum of a certain objective function. Notwithstanding some controversies regarding the novelty and the naming of the method, ELM has proven to be very useful for training single and multiple hidden layer feedforward neural networks.
Experimental results in classification
The purpose of this section is to compare the classification performance of an HMLP-EL, trained using the strategies described in the previous section, with the ones of similar models from the recent literature, namely:
- 1.
Dendritic morphological neural network trained by stochastic gradient descent (DMNN) (Zamora & Sossa, 2017);
- 2.
Dilation/erosion linear perceptron (DELP) (Araújo, Oliveira, & Meira, 2012);
- 3.
Morphological/linear neural network (MLNN) (Hernández et al., 2018);
- 4.
Morphological perceptron
Concluding remarks
In this paper, we introduced an artificial neural network model called hybrid morphological/ linear perceptron that is equipped with classical neurons as well as morphological units. Each of these morphological units or components computes the infimum of two elementary morphological operators, namely an erosion and an anti-dilation . Computational units of this type were already employed in previous models of morphological perceptrons that were trained using constructive algorithms (
Acknowledgments
This work was supported in part by CNPq under Grant No. 313145/2017-2 and by FAPESP under Grant Nos. 2018/13657-1 and 2017/10224-4. We would like to thank the anonymous reviewers, whose comments and suggestions helped to improve the paper.
References (57)
- et al.
A morphological-rank-linear evolutionary method for stock market prediction
Information Sciences
(2013) - et al.
A morphological neural network for binary classification problems
Engineering Applications of Artificial Intelligence
(2017) - et al.
Differential evolution training algorithm for dendrite morphological neural networks
Applied Soft Computing
(2018) - et al.
Decomposition of mappings between complete lattices by mathematical morphology, part 1. general lattices
Signal Processing
(1993) - et al.
Robust path-based spectral clustering
Pattern Recognition
(2008) - et al.
The algebraic basis of mathematical morphology: I. Dilations and erosions
Computer Vision, Graphics, and Image Processing
(1990) Approximation capabilities of multilayer feedforward networks
Neural Networks
(1991)- et al.
Extreme learning machine: Theory and applications
Neurocomputing
(2006) - et al.
Fuzzy lattice reasoning (FLR) classifier and its application for ambient ozone estimation
International Journal of Approximate Reasoning
(2007) - et al.
Fuzzy lattice neurocomputing (FLN) models
Neural Networks
(2000)
Neural networks with hybrid morphological/rank/linear nodes: a unifying framework with applications to handwritten character recognition
Pattern Recognition
Morphological bidirectional associative memories
Neural Networks
Why mathematical morphology needs complete lattices
Signal Processing
Deep learning in neural networks
Neural Networks
Efficient training for dendrite morphological neural networks
Neurocomputing
Lattice fuzzy transforms from the perspective of mathematical morphology
Fuzzy Sets and Systems
Morphological perceptrons with competitive learning: lattice-theoretical framework and constructive learning algorithm
Information Sciences
Dendrite morphological neurons trained by stochastic gradient descent
Neurocomputing
A morphological perceptron with gradient-based learning for Brazilian stock market forecasting
Neural Networks
A dilation-erosion-linear perceptron for Bovespa index prediction
Minimal representations for translation-invariant set mappings by mathematical morphology
SIAM Journal on Applied Mathematics
The sample complexity of pattern classification with neural networks: the size of the weights is more important than the size of the network
IEEE Transactions on Information Theory
Random search for hyper-parameter optimization
Journal of Machine Learning Research
Lattice theory
Neural networks for pattern recognition
Pattern recognition and machine learning (information science and statistics)
HMLP-EL
Cited by (31)
A multilayered bidirectional associative memory model for learning nonlinear tasks
2023, Neural NetworksGeneralized morphological components based on interval descriptors and n-ary aggregation functions
2022, Information SciencesSmooth dendrite morphological neurons
2021, Neural NetworksCitation Excerpt :Both approaches are trained with SGD, and they have the capability of extracting features. The Hybrid Morphological/Linear Perceptron (HMLP), proposed by Sussner and Campiotti (2020), is a two-layer model whose hidden layer includes morphological units and classical semi-linear neurons, and linear nodes in the output layer. HMLP is trained via an extreme learning machine approach.