Banchar Arnonkijpanich

Local matrix adaptation in topographic neural maps

The self-organizing map (SOM) and neural gas (NG) and generalizations thereof such as the generative topographic map constitute popular algorithms to represent data by means of prototypes arranged on a (hopefully) topology representing map. Most standard methods rely on the Euclidean metric, hence the resulting clusters tend to have isotropic form and they cannot account for local distortions or correlations of data. For this reason, several proposals exist in the literature which extend prototype-based clustering towards more general models which, for example, incorporate local principal directions into the winner computation. This allows to represent data faithfully using less prototypes. In this contribution, we establish a link of models which rely on local principal components (PCA), matrix learning, and a formal cost function of NG and SOM which allows to show convergence of the algorithm. For this purpose, we consider an extension of prototype-based clustering algorithms such as NG and SOM towards a more general metric which is given by a full adaptive matrix such that ellipsoidal clusters are accounted for. The approach is derived from a natural extension of the standard cost functions of NG and SOM (in the form of Heskes). We obtain batch optimization learning rules for prototype and matrix adaptation based on these generalized cost functions and we show convergence of the algorithm. The batch optimization schemes can be interpreted as local principal component analysis (PCA) and the local eigenvectors correspond to the main axes of the ellipsoidal clusters. Thus, this approach provides a cost function associated to proposals in the literature which combine SOM or NG with local PCA models. We demonstrate the behavior of matrix NG and SOM in several benchmark examples and in an application to image compression.

An approach based on digital image analysis to estimate the live weights of pigs in farm environments

We develop a machine vision-based method to estimate the live weight of pigs.The proposed approach provides circumstance-free image processing.It is practical to apply on farms without interrupting the routine life of pigs. In this study, an estimation system for the live weights of pigs is proposed that could be practically employed in a real farm environment without disturbing the animals. This approach is based on computer-assisted visual image capture and a supervised learning algorithm known as vector-quantized temporal associative memory (VQTAM). The method is composed of three parts, which are boundary detection, feature extraction, and pattern recognition. To identify an images edge, a method that is based on user interaction via mouse-clicking on the pig image is employed to avoid edge detection errors if the pigs image and its background are not in contrast. Two image features, (1) the average distance from the pigs centroid to the boundary points and (2) the pigs perimeter length, are extracted and used as the inputs of VQTAM. Next, the solutions from VQTAM are improved by an autoregressive model (AR) and locally linear embedding (LLE). This approach has been examined using a specific farm for a case study. The results indicate that the method based on VQTAM and improved by LLE provides the most accurate prediction with an error rate of less than 3% on average.

2010 Special Issue: Local matrix learning in clustering and applications for manifold visualization

Electronic data sets are increasing rapidly with respect to both, size of the data sets and data resolution, i.e. dimensionality, such that adequate data inspection and data visualization have become central issues of data mining. In this article, we present an extension of classical clustering schemes by local matrix adaptation, which allows a better representation of data by means of clusters with an arbitrary spherical shape. Unlike previous proposals, the method is derived from a global cost function. The focus of this article is to demonstrate the applicability of this matrix clustering scheme to low-dimensional data embedding for data inspection. The proposed method is based on matrix learning for neural gas and manifold charting. This provides an explicit mapping of a given high-dimensional data space to low dimensionality. We demonstrate the usefulness of this method for data inspection and manifold visualization.

PFP-PCA: Parallel Fixed Point PCA Face Recognition

With a high computational complexity of Eigenvector/Eigenvalue calculation, especially with a large database, of a traditional face recognition system, PCA, this paper proposes an alternative approach to utilize a fixed point algorithm for EVD stage optimization. We also proposed the optimization to reduce the complexity during the high computation stage, covariance matrix manipulation. In addition, the feasibility to enhance the speed-up over a single-core computation, parallelism, was investigated on the huge matrix calculation on both grayscale and RGB images. This mechanism, the so-called Parallel Fixed Point PCA (PFP-PCA), results in higher accuracy and lower complexity comparing to the traditional PCA leading to a high speed face recognition system.

Minimizing the total cost of hen allocation to poultry farms using hybrid Growing Neural Gas approach

The traditional Growing Neural Gas (GNG) was used for clustering hen houses by considering distance only.To improve the solutions, we developed the hybrid GNG by considering both distance and weights of hen house sizes.Routes determination to allocate hens to the hen houses was also carried out in order to minimize the total distance.The maximum allowable difference of ages of hens in the same hen house is considered as occurred in real practices.We proposed the multi-time period clustering based on chick ordering and hen house capacities. In this paper a decision support system to solve the problem of hen allocation to hen houses with the aim of minimizing the total cost is described. The total cost consists of farm utilization cost, hen transportation cost, and loss from mixing hens at different ages in the same hen houses. Clustering of hen houses using the traditional Growing Neural Gas (GNG) was first determined to allocate hens to the hen houses effectively. However, the traditional GNG often solves the clustering problem by considering distance only. Therefore the hybrid Growing Neural Gas (hGNG) considering both the distance from the centroids of the clusters to the hen houses and the weights of hen house sizes was proposed to solve the problem. In the second phase, allocating and determining routes to allocate hens to the hen houses using the nearest neighbor approach were carried out in order to minimize the total distance. The performance of the algorithm was measured using the relative improvement (RI), which compares the total costs of the hGNG and GNG algorithms and the current practice. The results obtained from this study show that the hGNG algorithm provides better total cost values than the firms current practice from 7.92% to 20.83%, and from 5.90% to 17.91% better than the traditional GNG algorithm. The results also demonstrate that the proposed method is useful not only for reducing the total cost, but also for efficient management of a poultry production system. Furthermore, the method used in this research should prove beneficial to other similar agro-food sectors in Thailand and around the world.

Matrix Learning for Topographic Neural Maps

The self-organizing map (SOM) and neural gas (NG) constitute popular algorithms to represent data by means of prototypes arranged on a topographic map. Both methods rely on the Euclidean metric, hence clusters are isotropic. In this contribution, we extend prototype-based clustering algorithms such as NG and SOM towards a metric which is given by a full adaptive matrix such that ellipsoidal clusters are accounted for. We derive batch optimization learning rules for prototype and matrix adaptation based on a general cost function for NG and SOM and we show convergence of the algorithm. It can be seen that matrix learning implicitly performs minor local principal component analysis (PCA) and the local eigenvectors correspond to the main axes of the ellipsoidal clusters. We demonstrate the behavior in several examples.

An artificial intelligence model to estimate the fat addition ratio for the mixing process in the animal feed industry

In animal feed pellets, the fat content is obtained either from the feed ingredients or is directly added during processing. Additional fat is required when the fat level in the feed ingredients is less than the desired level. This fat can be added either during the mixing process or after the pelleting process. However, adding fat at different time leads to different results. The addition of an increasing amount of fat during the mixing process decreases the pellet durability but enhances the pellet production rate. To avoid a reduction in the pellet durability, limiting the inclusion of fats in the mixer is suggested. The use of suitable fat addition ratios during mixing and after pelleting can improve the pellet quality and the production capability. Many factors significantly affect the decision of how much fat to add, such as the fiber inclusion content in the feed formulation, pellet die size, required feed durability, total required fat, and required additional fat. Due to frequent changes in the feed mix, anticipating the suitable amount of fat addition during the mixing process becomes a cumbersome task for a mill. In this paper, a model for estimating the amount of fat required in the mixer for each feed formulation is proposed. The model is based on the local linear map (LLM) and the back-propagation neural network (BPNN) methods. The LLM is used to identify which feed formulations require the addition of fat both during mixing and after pelleting, whereas the BPNN is employed for estimating the proper total fat required in the mixer, and the ratio of fat to add during the mixing process is subsequently estimated by subtracting the fat in the raw material from the total fat required in the mixer. The model is developed using data from one the largest feed mills in Thailand. The proposed model provides an accurate prediction and is practical for implementation in the mill that was studied.

Hybrid Balancing Technique Using GRSOM and Bootstrap Algorithms for Classifiers with Imbalanced Data

To deal with imbalanced data, this paper proposes a hybrid data balancing technique which incorporates both over and under-sampling approaches. This technique determines how much minority data should be grown as well as how much majority data should be reduced. In this manner, noise introduced to the data due to excessive over-sampling could be avoided. On top of that, the proposed data balancing technique helps to determine the appropriate size of the balanced data and thus computation time required for construction of classifiers would be more efficient. The data balancing technique over samples the minority data through GRSOM method and then under samples the majority data using the bootstrap sampling approach. GRSOM is used in this study because it grows new samples in a non-linear fashion and preserves the original data structure. Performance of the proposed method is tested using four data sets from UCI Machine Learning Repository. Once the data sets are balanced, the committee of classifiers is constructed using these balanced data. The experimental results reveal that our proposed data balancing method provides the best performance.

Image analysis for pig recognition based on size and weight

Stockman or farmers always have difficulty recognition of pig mass in their farms. The typical approach is to approximate from age of pigs, daily-given feed, or from experience of human vision. Another practical approach to instantly measure mass of pigs is to use machine vision. The objective of this paper is to use a developed machine vision to analyze pig mass for detection of size and weight of pigs in farm. The pig mass is processed from physical features captured from digital image and their liveweights are approximated from artificial neural network. This neural network model is based on vector-quantized temporal associative memory (VQTAM) and locally linear embedding (LLE). The elementary results showed that the mass approximation of pig weight had acceptable accuracy and it was practical in pig farms.

Global Coordination Based on Matrix Neural Gas for Dynamic Texture Synthesis

Matrix neural gas has been proposed as a mathematically well-founded extension of neural gas networks to represent data in terms of prototypes and local principal components in a smooth way. The additional information provided by local principal directions can directly be combined with charting techniques such that a nonlinear embedding of a data manifold into low dimensions results for which an explicit function as well as an approximate inverse exists. In this paper, we show that these ingredients can be used to embed dynamic textures in low dimensional spaces such that, together with a traversing technique in the low dimensional representation, efficient dynamic texture synthesis can be obtained.