Benchmarking Least Squares Support Vector Machine Classifiers

In Support Vector Machines (SVMs), the solution of the classification problem is characterized by a (convex) quadratic programming (QP) problem. In a modified version of SVMs, called Least Squares SVM classifiers (LS-SVMs), a least squares cost function is proposed so as to obtain a linear set of equations in the dual space. While the SVM classifier has a large margin interpretation, the LS-SVM formulation is related in this paper to a ridge regression approach for classification with binary targets and to Fisher's linear discriminant analysis in the feature space. Multiclass categorization problems are represented by a set of binary classifiers using different output coding schemes. While regularization is used to control the effective number of parameters of the LS-SVM classifier, the sparseness property of SVMs is lost due to the choice of the 2-norm. Sparseness can be imposed in a second stage by gradually pruning the support value spectrum and optimizing the hyperparameters during the sparse approximation procedure. In this paper, twenty public domain benchmark datasets are used to evaluate the test set performance of LS-SVM classifiers with linear, polynomial and radial basis function (RBF) kernels. Both the SVM and LS-SVM classifier with RBF kernel in combination with standard cross-validation procedures for hyperparameter selection achieve comparable test set performances. These SVM and LS-SVM performances are consistently very good when compared to a variety of methods described in the literature including decision tree based algorithms, statistical algorithms and instance based learning methods. We show on ten UCI datasets that the LS-SVM sparse approximation procedure can be successfully applied.     

The effect of random grain distributions on fatigue crack initiation in a notched coarse grained superalloy specimen

Coarse grained superalloys are of large interest in high temperature applications, and can be found in e.g. gas turbine components, where great care must be given with respect to high temperature fatigue. Due to the large grain size, the material behaviour at e.g. sharp notches cannot be considered homogeneous. As a consequence, the fatigue behaviour is likely to expose a large variation. In order to numerically investigate this variation, a Monte Carlo analysis has been carried out by 100 FE-simulations of notched specimens, where placements and orientations of the grains were randomised. Furthermore, each grain was modelled as a unique single-crystal, displaying both anisotropic elastic and plastic behaviour and tension/compression asymmetry. The effect of randomness was investigated by the obtained dispersion in fatigue crack initiation life. It was concluded that the fatigue life behaviour of coarse grained nickel-base superalloys may show a considerable variation, which cannot be captured by one single deterministic analysis based on data for a homogenised material. Furthermore, the dispersion is of such a magnitude that it needs to be taken into account in industrial applications where highly stressed coarse grained materials are used.     

Natural Language and Logic of Agency

This light piece reflects on analogies between two often disjoint streams of research: the logical semantics and pragmatics of natural language and dynamic logics of general information-driven agency. The two areas show significant overlap in themes and tools, and yet, the focus seems subtly different in each, defying a simple comparison. We discuss some unusual questions that emerge when the two are put side by side, without any pretense at covering the whole literature or at reaching definitive conclusions.     

Escape trajectories are deflected when fish larvae intercept their own C-start wake

Fish larvae may intercept their own wake during sharp turns, which might affect their escape performance. We analysed C-starts of larval zebrafish (Danio rerio, Hamilton, 1822) using a computational fluid dynamics approach that simulates free swimming (swimming trajectory is determined by fluid forces) by coupling hydrodynamics and body dynamics. The simulations show that fish may intercept their own wake when they turn by 100–180°. During stage 1 of a C-start, the fish generates a strong jet at the tail that is shed into the wake. During stage 2, the fish intercepts this wake. Counterfactual simulations showed that wake interception increased the lateral force on the fish and reduced the fish''s turning angle by more than 5°. Wake interception caused no significant acceleration tangential to the trajectory of the fish and did not affect total power output. While experimental and simulation evidence suggests that fish larvae can either undershoot or intercept but not overshoot their wake, our simulations show that larger fish might be able to avoid intercepting their wake by either under- or overshooting. As intercepting its own wake modifies the fish''s escape trajectory, fish should account for this effect when planning their escape route.     

Separation of L‐aspartic acid and L‐glutamic acid mixtures for use in the production of bio‐based chemicals

BACKGROUND: Amino acids are promising feedstocks for the chemical industry due to their chemical functionality. They can be obtained by the hydrolysis of potentially inexpensive protein streams such as the byproduct of biofuel production. However, individual amino acids are required before they can be used for the further production of chemicals. Here, the separation of L‐aspartic acid (Asp) and L‐glutamic acid (Glu) mixture, which can be isolated from protein hydrolysis solutions at low pH or from electrodialysis of complex amino acid mixtures, was studied. RESULTS: Glu was converted into L‐pyroglutamic acid (pGlu) which can be separated from the mixture of Asp and Glu due to its higher solubility in water. The conversion was carried out under aqueous or melt conditions. Under aqueous conditions, the conversion was studied as a factor of time, temperature and the amount of Glu. The conversion was specific with high yield and not effected by Asp. After pGlu was separated from Asp and residual Glu by solubility difference, it can be transferred back to Glu through hydrolysis. CONCLUSION: The conversion of Glu to pGlu is specific and can be applied to separation Asp and Glu for their use in the production of bio‐based chemicals. Copyright © 2012 Society of Chemical Industry     

Robust Facility Location Problem for Hazardous Waste Transportation

We consider a robust facility location problem for hazardous materials (hazmat) transportation considering routing decisions of hazmat carriers. Given a network and a known set of nodes from which hazmat originate, we compute the locations of hazmat processing sites (e.g. incinerators) which will minimize total cost, in terms of fixed facility cost, transportation cost, and exposure risk. We assume that hazmat will be taken to the closest existing processing site. We present an exact full enumeration method, which is useful for small or medium-size problems. For larger problems, the use of a genetic algorithm is explored. Through numerical experiments, we discuss the impact of uncertainty and robust optimization in the hazmat combined location-routing problem.     

Degradation of erythromycin in honey and selection of suitable marker residues for food safety analysis

Extra-label use of veterinary medications in apiculture is a practice which is known to result in the presence of drug residues in honey. Erythromycin has been used by some beekeepers in an attempt to control foulbrood diseases in honey bees. It is recognised that erythromycin degrades in acidic aqueous solutions to yield predominantly anhydroerythromycin. Honey is an acidic medium in which erythromycin should also degrade. Three degradation products with a molecular weight of 715 Da (anhydroerythromycin, erythromycin enol ether, and an unidentified but suspected related isomer) have been confirmed to be formed in honey. Erythromycin was found to degrade rapidly in honey at typical hive temperature with a half-life of less than one day. Based on these results, it is recommended that food safety laboratories include degradation products in analytical testing protocols when examining honey for erythromycin residues.     

Symmetric Assumptions in the Theory of Disruptive Innovation: Theoretical and Managerial Implications

The literature on disruptive innovation has convincingly explained why many established firms encounter problems under conditions of discontinuous change. Incumbents fail to invest in new technologies that are not demanded by their existing customers. This argument is grounded in resource dependency theory and the associated assumption that existing customers control a firm's internal resource allocation processes. While the problem of disruptive innovation has been convincingly explained, there is still a need for managerial solutions. We argue that a key reason why such solutions are lacking can be found in the asymmetric assumptions made in the original theory of disruptive innovation. Specifically, we identify two related forms of asymmetry. First, the focal (incumbent) firm is treated as a collection of heterogeneous actors with different preferences, incentives and competencies, whereas firms in the surrounding environment are treated as if they contained no such heterogeneity. Second, the theory of disruptive innovation describes incumbents as controlled by their environment, but has failed to recognize that the environment can also be influenced. In this paper we argue that a more symmetric theory of disruptive innovation – i.e. one that treats all similar entities in the same way – opens up for a range of interesting managerial solutions.