Computational study of the step size parameter of the subgradient optimization method

The subgradient optimization method is a simple and flexible linear programming iterative algorithm. It is much simpler than Newton's method and can be applied to a wider variety of problems. It also converges when the objective function is non-differentiable. Since an efficient algorithm will not only produce a good solution but also take less computing time, we always prefer a simpler algorithm with high quality. In this study a series of step size parameters in the subgradient equation is studied. The performance is compared for a general piecewise function and a specific p-median problem. We examine how the quality of solution changes by setting five forms of step size parameter.

Convex Hulls in Image Processing : A Scoping Review

The demands of image processing related systems are robustness, high recognition rates, capability to handle incomplete digital information, and magnanimous flexibility in capturing shape of an object in an image. It is exactly here that, the role of convex hulls comes to play. The objective of this paper is twofold. First, we summarize the state of the art in computational convex hull development for researchers interested in using convex hull image processing to build their intuition, or generate nontrivial models. Secondly, we present several applications involving convex hulls in image processing related tasks. By this, we have striven to show researchers the rich and varied set of applications they can contribute to. This paper also makes a humble effort to enthuse prospective researchers in this area. We hope that the resulting awareness will result in new advances for specific image recognition applications.

Density based fuzzy c-means clustering of non-convex patterns

We propose a new technique to perform unsupervised data classification (clustering) based on density induced metric and non-smooth optimization. Our goal is to automatically recognize multidimensional clusters of non-convex shape. We present a modification of the fuzzy c-means algorithm, which uses the data induced metric, defined with the help of Delaunay triangulation. We detail computation of the distances in such a metric using graph algorithms. To find optimal positions of cluster prototypes we employ the discrete gradient method of non-smooth optimization. The new clustering method is capable to identify non-convex overlapped d-dimensional clusters.

Fuzzy clustering of non-convex patterns using global optimization

This paper discusses various extensions of the classical within-group sum of squared errors functional, routinely used as the clustering criterion. Fuzzy c-means algorithm is extended to the case when clusters have irregular shapes, by representing the clusters with more than one prototype. The resulting minimization problem is non-convex and non-smooth. A recently developed cutting angle method of global optimization is applied to this difficult problem

Convex hulls as an hypothesis language bias

Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces which biases such systems towards particular hypothesis types. However, there is reason to believe that many domains have underlying concepts which do not involve axis orthogonal surfaces. Further, the multiplicity of small decision regions mitigates against any holistic appreciation of the theories produced by these systems, notwithstanding the fact that many of the small regions are individually comprehensible. We propose the use of less strongly biased hypothesis languages which might be expected to model' concepts using a number of structures close to the number of actual structures in the domain. An instantiation of such a language, a convex hull based classifier, CHI, has been implemented to investigate modeling concepts as a small number of large geometric structures in n-dimensional space. A comparison of the number of regions induced is made against other well-known systems on a representative selection of largely or wholly continuous valued machine learning tasks. The convex hull system is shown to produce a number of induced regions about an order of magnitude less than well-known systems and very close to the number of actual concepts. This representation, as convex hulls, allows the possibility of extraction of higher level mathematical descriptions of the induced concepts, using the techniques of computational geometry.

Learning from large data : bias, variance, sampling, and learning curves

One of the fundamental machine learning tasks is that of predictive classification. Given that organisations collect an ever increasing amount of data, predictive classification methods must be able to effectively and efficiently handle large amounts of data. However, it is understood that present requirements push existing algorithms to, and sometimes beyond, their limits since many classification prediction algorithms were designed when currently common data set sizes were beyond imagination. This has led to a significant amount of research into ways of making classification learning algorithms more effective and efficient. Although substantial progress has been made, a number of key questions have not been answered. This dissertation investigates two of these key questions. The first is whether different types of algorithms to those currently employed are required when using large data sets. This is answered by analysis of the way in which the bias plus variance decomposition of predictive classification error changes as training set size is increased. Experiments find that larger training sets require different types of algorithms to those currently used. Some insight into the characteristics of suitable algorithms is provided, and this may provide some direction for the development of future classification prediction algorithms which are specifically designed for use with large data sets. The second question investigated is that of the role of sampling in machine learning with large data sets. Sampling has long been used as a means of avoiding the need to scale up algorithms to suit the size of the data set by scaling down the size of the data sets to suit the algorithm. However, the costs of performing sampling have not been widely explored. Two popular sampling methods are compared with learning from all available data in terms of predictive accuracy, model complexity, and execution time. The comparison shows that sub-sampling generally products models with accuracy close to, and sometimes greater than, that obtainable from learning with all available data. This result suggests that it may be possible to develop algorithms that take advantage of the sub-sampling methodology to reduce the time required to infer a model while sacrificing little if any accuracy. Methods of improving effective and efficient learning via sampling are also investigated, and now sampling methodologies proposed. These methodologies include using a varying-proportion of instances to determine the next inference step and using a statistical calculation at each inference step to determine sufficient sample size. Experiments show that using a statistical calculation of sample size can not only substantially reduce execution time but can do so with only a small loss, and occasional gain, in accuracy. One of the common uses of sampling is in the construction of learning curves. Learning curves are often used to attempt to determine the optimal training size which will maximally reduce execution time while nut being detrimental to accuracy. An analysis of the performance of methods for detection of convergence of learning curves is performed, with the focus of the analysis on methods that calculate the gradient, of the tangent to the curve. Given that such methods can be susceptible to local accuracy plateaus, an investigation into the frequency of local plateaus is also performed. It is shown that local accuracy plateaus are a common occurrence, and that ensuring a small loss of accuracy often results in greater computational cost than learning from all available data. These results cast doubt over the applicability of gradient of tangent methods for detecting convergence, and of the viability of learning curves for reducing execution time in general.

Convex hulls in concept induction

Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces. This strongly biases such systems towards particular hypothesis types but there is reason believe that many domains have underlying concepts which do not involve axis orthogonal surfaces. Further, the multiplicity of small decision regions mitigates against any holistic appreciation of the theories produced by these systems, notwithstanding the fact that many of the small regions are individually comprehensible. This thesis investigates modeling concepts as large geometric structures in n-dimensional space. Convex hulls are a superset of the set of axis orthogonal hyperrectangles into which axis orthogonal systems partition the instance space. In consequence, there is reason to believe that convex hulls might provide a more flexible and general learning bias than axis orthogonal regions. The formation of convex hulls around a group of points of the same class is shown to be a usable generalisation and is more general than generalisations produced by axis-orthogonal based classifiers, without constructive induction, like decision trees, decision lists and rules. The use of a small number of large hulls as a concept representation is shown to provide classification performance which can be better than that of classifiers which use a large number of small fragmentary regions for each concept. A convex hull based classifier, CH1, has been implemented and tested. CH1 can handle categorical and continuous data. Algorithms for two basic generalisation operations on hulls, inflation and facet deletion, are presented. The two operations are shown to improve the accuracy of the classifier and provide moderate classification accuracy over a representative selection of typical, largely or wholly continuous valued machine learning tasks. The classifier exhibits superior performance to well-known axis-orthogonal-based classifiers when presented with domains where the underlying decision surfaces are not axis parallel. The strengths and weaknesses of the system are identified. One particular advantage is the ability of the system to model domains with approximately the same number of structures as there are underlying concepts. This leads to the possibility of extraction of higher level mathematical descriptions of the induced concepts, using the techniques of computational geometry, which is not possible from a multiplicity of small regions.

Construction of k-Lipschitz triangular norms and conorms from empirical data

This paper examines the practical construction of k-Lipschitz triangular norms and conorms from empirical data. We apply a characterization of such functions based on k-convex additive generators and translate k-convexity of piecewise linear strictly decreasing functions into a simple set of linear inequalities on their coefficients. This is the basis of a simple linear spline-fitting algorithm, which guarantees k-Lipschitz property of the resulting triangular norms and conorms.

Monitoring the corrosion of concrete reinforcement using control curves

Corrosion testing (half-cell and LPR) was carried out on a number reinforced concrete panels which had been taken from the fascia of a twenty five year old high rise building in Melbourne, Australia. Corrosion, predominantly as a result of carbonation of the concrete, was associated with a limited amount of cracking. A monitoring technique was established in which probe electrodes (reference and counter) were retro-fitted into the concrete. The probe electrode setup was identical for all panels tested. It was found that the corrosion behaviour of all panels tested closely fitted a family of results when the corrosion potential is plotted against the polarisation resistance (Rp). This enabled the development of a so-called 'control curve' relating the corrosion potential to the Rp for all of the panels under investigation. This relationship was also confirmed on laboratory samples, indicating that for a fixed geometry and experimental conditions a relationship between the potential and polarisation resistance of steel can be established for the steel-concrete system. Experimental results will be presented which indicate that for a given monitoring cell geometry, it may be possible to propose criteria for the point at which remediation measures should be considered. The establishment of such a control curve has enabled the development of a powerful monitoring tool for the assessment of a number of proposed corrosion remediation techniques. The actual effect of any corrosion remediation technique becomes clearly apparent via the type and magnitude of deviation of post remediation data from the original (preremediation) control curve.

Effects of replication on measured S-N curves of metals

Replication, or repeated tests at the same stress amplitude, is used to provide statistical confidence in life data during the development of S-N curves. This paper discusses the effects of replication on the measurement of S-N curves and presents an alternative to traditional replication methods for the determination of S-N curves, particularly for the development of preliminary S-N curves. Using specimens made out of the extruded bars of a magnesium alloy, it is demonstrated that the S-N curve estimated using the data from non-replication tests is almost same as that from replication tests. The advantage of using non-replication fatigue tests is that it uses fewer specimens, in this instance, only half of that required for 50% replication fatigue test, to achieve the same estimation as that of the replication fatigue tests. Another advantage of using non-replication fatigue tests is that it can detect the non-linearity using limited specimens.

Software reliability growth models based on local polynomial modeling with kernel smoothing

Software reliability growth models (SRGMs) are extensively employed in software engineering to assess the reliability of software before their release for operational use. These models are usually parametric functions obtained by statistically fitting parametric curves, using Maximum Likelihood estimation or Least–squared method, to the plots of the cumulative number of failures observed N(t) against a period of systematic testing time t. Since the 1970s, a very large number of SRGMs have been proposed in the reliability and software engineering literature and these are often very complex, reflecting the involved testing regime that often took place during the software development process. In this paper we extend some of our previous work by adopting a nonparametric approach to SRGM modeling based on local polynomial modeling with kernel smoothing. These models require very few assumptions, thereby facilitating the estimation process and also rendering them more relevant under a wide variety of situations. Finally, we provide numerical examples where these models will be evaluated and compared.

Effects of replication on measured S-N curves of metals

Replication, or repeated tests at the same stress amplitude, is used to provide statistical confidence in life data during the development of S-N curves. This paper discusses the effects of replication on the measurement of S-N curves and presents an alternative to traditional replication methods for the determination of S-N curves, particularly for the development of preliminary S-N curves. Using specimens made out of the extruded bars of a magnesium alloy, it is demonstrated that the S-N curve estimated using the data from non-replication tests is almost same as that from replication tests. The advantage of using non-replication fatigue tests is that it uses fewer specimens, in this instance, only half of that required for 50% replication fatigue test, to achieve the same estimation as that of the replication fatigue tests. Another advantage of using non-replication fatigue tests is that it can detect the non-linearity using limited specimens.