China
Africa
Explore chapters and articles related to this topic
Artificial Neural Networks
Published in Paresh Chra Deka, A Primer on Machine Learning Applications in Civil Engineering, 2019
Unlike the supervised networks, in an unsupervised neural network the training of data sets is carried out without the supervision of a teacher. That means the learning process in unsupervised neural networks is carried out by a self-organizing pattern. During the training process, no external reference is used to adjust the weight of the network. The input vectors of similar patterns will try to form a cluster in a training process. When a new input is defined, then the neural network of unsupervised learning will produce an output based on the pattern to which it belongs. The correct output will not be available during the course of training. A typical unsupervised network consists of an input and a competitive layer. The neurons in the competitive layer will match with each other using some competitive rules during the learning period and produce the best output for the given input pattern. Self-organizing map (SOM) and Adaptive Resonance Theory (ART) are the most widely used neural networks in unsupervised learning algorithms. Unsupervised learning can be further classified into clustering and association problems. A clustering problem will create inherent grouping in the data based on its behavior. While in an association rule of learning, it will discover rules to describe the data, such as ‘people who buy X also tend to buy Y.’ k-means for clustering problems and a priori algorithms for association problems are some popular algorithms for unsupervised learning.
Dimensionality Reduction — Nonlinear Methods
Published in Wendy L. Martinez, Angel R. Martinez, Jeffrey L. Solka, Exploratory Data Analysis with MATLAB®, 2017
Wendy L. Martinez, Angel R. Martinez, Jeffrey L. Solka
The self-organizing map (SOM) is a tool for the exploration and visualization of high-dimensional data [Kohonen, 1998]. It derives an orderly mapping of data onto a regular, low-dimensional grid. The dimensionality of the grid is usually d = 2 for ease of visualization. It converts complex, nonlinear relationships in the high-dimensional space into simpler geometric relationships such that the important topological and metric relationships are conveyed. The data are organized on the grid in such a way that observations that are close together in the high-dimensional space are also closer to each other on the grid. Thus, this is very similar to the ideas of MDS, except that the positions in the low-dimensional space are restricted to the grid. The grid locations are denoted by ri.
Computational material design of filled rubbers using multi-objective design exploration
Published in Alexander Lion, Michael Johlitz, Constitutive Models for Rubber X, 2017
M. Koishi, N. Kowatari, B. Figliuzzi, M. Faessel, F. Willot, D. Jeulin
A sophisticated visualization technique is required to show high-dimensional objective functions and design parameters simultaneously. To visualize high-dimensional data, self-organizing map (SOM) proposed by Kohonen (1995) is employed in this work. SOM is one of the neural network model based on unsupervised and competitive learning. It provides a mapping with preserving topology from the high-dimensional space to two-dimensional plane, so called map. Although SOM does not remain the direction and distance in the original high-dimensional space, nearby points in the high-dimensional space are mapped to the nearby points on SOM. Roughly speaking, a relation between high-dimensional data and SOM is similar to the relation between the earth and a world map. Nearby countries on the earth are mapped to nearby positions on a world map. SOM is useful not only for visualization of highdimensional data but also for the cluster analysis for design problems in industry (Koishi et al. 2006 & Koishi et al. 2014).
Elucidating the prediction capability of neural network model for estimation of crop water stress index of rice
Published in ISH Journal of Hydraulic Engineering, 2023
Aschalew Cherie Workneh, K.S. Hari Prasad, Chandra Shekhar Ojha
The self-organizing map, commonly known as the SOM feature map or Kohonen map, is a widely used unsupervised ANN method for clustering massive amounts of data, combining various datasets to find patterns, and creating powerful visualizations in any discipline (Kohonen 1982). It has the distinct ability to convert high-dimensional input data to low-dimensional output data (two-dimensional). To put it another way, it can arrange and organize massive collections of data into a two-dimensional array, reducing their dimensionality (Figure 5). The clustering approach may also turn complex nonlinear relationships into simple ones (Kohonen 1982; Kumar et al. 2021). Map units (neurons) used the principle of fire together or wire together (compete with each other) in unsupervised learning until they had adjusted to the input signal patterns. Similar patterns by the same output neuron or an adjacent one are represented by clustered input data (Stefanovič and Kurasova 2011).
Application of the self-organizing map in the classification of natural antioxidants in commercial biodiesel
Published in Biofuels, 2021
Marissa Kimura, Felipe Y. Savada, Daniele L.M. Tashima, Érica S. Romagnoli, Letícia T. Chendynski, Livia R.C. Silva, Dionisio Borsato
There are several ways of analyzing data such as regressions, neural networks, principal component analysis, genetic algorithms, and others. The choice of the type of modeling to be done varies with the amount of data, whether it exhibits linear behavior or not, and whether there are many variables involved. Artificial neural networks were motivated by and developed based on the performance of the human brain to process information and perform perceptive task recognition [17]. One artificial neural network, proposed by Teuvo Kohonen in 1982, is the self-organizing map (SOM) that has unsupervised learning and allows the identification of unknown patterns [18–20].
Estimating Forest Fire Losses Using Stochastic Approach: Case Study of the Kroumiria Mountains (Northwestern Tunisia)
Published in Applied Artificial Intelligence, 2018
Ahmed Toujani, Hammadi Achour, Sami Faïz
The HMM with discrete observations is generally the most used since its robustness in training in comparison to continuous HMMs, where observations are formed by mixtures of Gaussians as opposed to discrete symbols (Rabiner 1989). Therefore, we used the Self Organized Map (SOM) to cluster the vector data of fire factors (Kohonen, Kaski, and Lappalainen 1997). The Self-organizing map is a nonparametric and nonlinear neural network that explores data using unsupervised learning. In the input layer, the neurons correspond to the variables describing the vector data. The output layer is most often organized as a grid of two-dimensional neurons.