Jan Verwaeren

Learning partial ordinal class memberships with kernel-based proportional odds models

As an extension of multi-class classification, machine learning algorithms have been proposed that are able to deal with situations in which the class labels are defined in a non-crisp way. Objects exhibit in that sense a degree of membership to several classes. In a similar setting, models are developed here for classification problems where an order relation is specified on the classes (i.e., non-crisp ordinal regression problems). As for traditional (crisp) ordinal regression problems, it is argued that the order relation on the classes should be reflected by the model structure as well as the performance measure used to evaluate the model. These arguments lead to a natural extension of the well-known proportional odds model for non-crisp ordinal regression problems, in which the underlying latent variable is not necessarily restricted to the class of linear models (by using kernel methods).

Supervised learning algorithms for multi-class classification problems with partial class memberships

In several application domains such as biology, computer vision, social network analysis and information retrieval, multi-class classification problems arise in which data instances not simply belong to one particular class, but exhibit a partial membership to several classes. Existing machine learning or fuzzy set approaches for representing this type of fuzzy information mainly focus on unsupervised methods. In contrast, we present in this article supervised learning algorithms for classification problems with partial class memberships, where class memberships instead of crisp class labels serve as input for fitting a model to the data. Using kernel logistic regression (KLR) as a baseline method, first a basic one-versus-all approach is proposed, by replacing the binary-coded label vectors with [0,1]-valued class memberships in the likelihood. Subsequently, we use this KLR extension as base classifier to construct one-versus-one decompositions, in which partial class memberships are transformed and estimated in a pairwise manner. Empirical results on synthetic data and a real-world application in bioinformatics confirm that our approach delivers promising results. The one-versus-all method yields the best computational efficiency, while the one-versus-one methods are preferred in terms of predictive performance, especially when the observed class memberships are heavily unbalanced.

Green leaf volatile production by plants: a meta‐analysis

666 I. Introduction 667 II. Biosynthesis 667 III. Meta-analysis 669 IV. The type of stress influences the total amount of GLVs released 669 V. Herbivores can modulate the wound-induced release of GLVs 669 VI. Fungal infection greatly induces GLV production 672 VII. Monocots and eudicots respond differentially to different types of stress 673 VIII. The type of stress does not influence the proportion of GLVs per chemical class 673 IX. The type of stress does influence the isomeric ratio within each chemical class 674 X. GLVs: from signal perception to signal transduction 676 XI. GLVs influence the C/N metabolism 677 XII. Interaction with plant hormones 678 XIII. General conclusions and unanswered questions 678 Acknowledgements 679 References 679 SUMMARY: Plants respond to stress by releasing biogenic volatile organic compounds (BVOCs). Green leaf volatiles (GLVs), which are abundantly produced across the plant kingdom, comprise an important group within the BVOCs. They can repel or attract herbivores and their natural enemies; and they can induce plant defences or prime plants for enhanced defence against herbivores and pathogens and can have direct toxic effects on bacteria and fungi. Unlike other volatiles, GLVs are released almost instantly upon mechanical damage and (a)biotic stress and could thus function as an immediate and informative signal for many organisms in the plants environment. We used a meta-analysis approach in which data from the literature on GLV production during biotic stress responses were compiled and interpreted. We identified that different types of attackers and feeding styles add a degree of complexity to the amount of emitted GLVs, compared with wounding alone. This meta-analysis illustrates that there is less variation in the GLV profile than we presumed, that pathogens induce more GLVs than insects and wounding, and that there are clear differences in GLV emission between monocots and dicots. Besides the meta-analysis, this review provides an update on recent insights into the perception and signalling of GLVs in plants.

Milk fatty acids as possible biomarkers to diagnose hyperketonemia in early lactation

The aim of this study was to assess the potential of milk fatty acids as diagnostic tool for hyperketonemia of 93 dairy cows in a 3×2 factorial arrangement. Cows were fed a glucogenic or lipogenic diet and originally were intended to be subjected to a 0-, 30-, or 60-d dry period. Nevertheless, some of the cows, which were intended for inclusion in the 0-d dry period group, dried off spontaneously. Milk was collected in wk 2, 3, 4, and 8 of lactation for milk fat analysis. Blood was sampled from wk 2 to 8 after parturition for β-hydroxybutyrate (BHBA) analysis. Cases were classified into 2 groups: hyperketonemia (BHBA ≥1.2mmol/L) and nonhyperketonemia (BHBA <1.2mmol/L). Concentrations of 45 milk fatty acids and ratios of anteiso C15:0-to-anteiso C17:0 and C18:1 cis-9-to-C15:0 were subjected to a logistic regression analysis (stepwise forward method). The milk fat C18:1 cis-9-to-C15:0 ratio revealed the most discriminating factor for diagnosis of hyperketonemia. Ninety percent of nonhyperketonemia cases showed a milk fat C18:1 cis-9-to-C15:0 ratio of 40 or lower, whereas 70% of cows suffering from hyperketonemia showed milk fat C18:1 cis-9-to-C15:0 ratios exceeding 40. Additionally, cows with a milk fat ratio C18:1 cis-9-to-C15:0 of at least 45 in wk 2 after parturition had about 50% chance to encounter blood plasma BHBA values of 1.2mmol/L or more during the first 8 wk of lactation. Of the cows not suffering from hyperketonemia during the first 2 mo of lactation, only 9% exceeded this wk 2 threshold. Practical implementation requires routine analysis of both milk fatty acids, which currently is lacking for C15:0. The inclusion of other variables, such as test-day information and a more frequent sampling protocol should be considered to further improve diagnostic performance of this biomarker.

A search grid for parameter optimization as a byproduct of model sensitivity analysis

Inverse problem solving, i.e. the retrieval of optimal values of model parameters from experimental data, remains a bottleneck for modelers. Therefore, a large variety of (heuristic) optimization algorithms has been developed to deal with the inverse problem. However, in some cases, the use of a grid search may be more appropriate or simply more practical. In this paper an approach is presented to improve the selection of the grid points to be evaluated and which does not depend on the knowledge or availability of the underlying model equations. It is suggested that using the information acquired through a sensitivity analysis can lead to better grid search results. Using the sensitivity analysis information, a Gauss-Newton-like matrix is constructed and the eigenvalues and eigenvectors of this matrix are employed to transform naive search grids into better thought-out ones. After a theoretical analysis of the approach, some computational experiments are performed using a simple linear model, as well as more complex nonlinear models.

Countering the negative search bias of ant colony optimization in subset selection problems

In their quest to find a good solution to a given optimization problem, metaheuristic search algorithms intend to explore the search space in a useful and efficient manner. Starting from an initial state or solution(s), they are supposed to evolve towards high-quality solutions. For some types of genetic algorithms (GAs), it has been shown that the population of chromosomes can converge to very bad solutions, even for trivial problems. These so-called deceptive effects have been studied intensively in the field of GAs and several solutions to these problems have been proposed. Recently, similar problems have been noticed for ant colony optimization (ACO) as well. As for GAs, ACOs search can get biased towards low-quality regions in the search space, probably resulting in bad solutions. Some methods have been proposed to investigate the presence and strength of this negative bias in ACO. We present a framework that is capable of eliminating the negative bias in subset selection problems. The basic Ant System algorithm is modified to make it more robust to the presence of negative bias. A profound simulation study indicates that the modified Ant System outperforms the original version in problems that are susceptible to bias. Additionally, the proposed methodology is incorporated in the Max-Min AS and applied to a real-life subset selection problem.

The digitization of a food package’s life cycle : existing and emerging computer systems in the pre-logistics phase

Abstract Traditional design and production methods for food packages become less and less suitable to rapidly respond to ever-changing requirements and regulations. Computer systems applied in discrete manufacturing (ranging from computer-aided-technologies to image analysis systems) are now also specifically developed for and gradually adopted by the food package industry to improve efficiency in terms of material usage, operational costs, and food loss, and to allow the development of more performant and sustainable food packages. In this paper, an extensive overview is provided of such systems that, when combined, offer the perspective to realize a more holistic research, design and production approach that fits within the spirit of the fourth industrial revolution. Special attention is given to the importance of information from and knowledge about the logistics and post-logistics phase of a food packages life cycle in the manufacturing process. The main purpose of this review paper is to provide, for the first time, a complete and coherent overview of the digitization of a food packages life cycle that can be used as a blueprint for future research, development and discussion in this emerging research topic.

Species prevalence and disease progression studies demonstrate a seasonal shift in the Alternaria population composition on potato

To assess the early blight/brown spot (EB/BS) incidence in Flanders (Belgium), a survey was set up consisting of potato fields in 22 locations which were monitored and scored over the duration of the growing seasons 2014 and 2015. The survey demonstrated that the average disease incidence in 2014 was higher than in 2015. Soil type, rainfall and temperature were additionally analysed during the infection process. In 2014, potato plants grown in sandy soils had more EB/BS disease than those that were grown in clay or loamy soils. However, due to the low disease incidence, the difference in disease levels observed between different soil types in 2014 could not be repeated in 2015. A window-pane analysis demonstrated that rainfall and humidity account for the differences in disease incidence between the two growing seasons. During the course of the survey, the species composition in symptomatic leaves was assessed using real-time PCR. Remarkably, small-spored Alternaria species, such as A. alternata and A. arborescens, rather than the more virulent A. solani were the predominant species on potato leaves throughout the growing season. As the disease progressed, the share of A. solani increased. In view of these results, the virulence of a collected set of Alternaria isolates was assessed by an in vitro assay. Besides A. solani being more virulent than A. alternata or A. arborescens, the most abundant species isolated from symptomatic potato leaves was A. arborescens. This article is protected by copyright. All rights reserved.

Congolese Rhizospheric Soils as a Rich Source of New Plant Growth-Promoting Endophytic Piriformospora Isolates

In the last decade, there has been an increasing focus on the implementation of plant growth-promoting (PGP) organisms as a sustainable option to compensate for poor soil fertility conditions in developing countries. Trap systems were used in an effort to isolate PGP fungi from rhizospheric soil samples collected in the region around Kisangani in the Democratic Republic of Congo. With sudangrass as a host, a highly conducive environment was created for sebacinalean chlamydospore formation inside the plant roots resulting in a collection of 51 axenically cultured isolates of the elusive genus Piriformospora (recently transferred to the genus Serendipita). Based on morphological data, ISSR fingerprinting profiles and marker gene sequences, we propose that these isolates together with Piriformospora williamsii constitute a species complex designated Piriformospora (= Serendipita) ‘williamsii.’ A selection of isolates strongly promoted plant growth of in vitro inoculated Arabidopsis seedlings, which was evidenced by an increase in shoot fresh weight and a strong stimulation of lateral root formation. This isolate collection provides unprecedented opportunities for fundamental as well as translational research on the Serendipitaceae, a family of fungal endophytes in full expansion.

Development and evaluation of land use regression models for black carbon based on bicycle and pedestrian measurements in the urban environment

Land use regression (LUR) modelling is increasingly used in epidemiological studies to predict air pollution exposure. The use of stationary measurements at a limited number of locations to build a LUR model, however, can lead to an overestimation of its predictive abilities. We use opportunistic mobile monitoring to gather data at a high spatial resolution to build LUR models to predict annual average concentrations of black carbon (BC). The models explain a significant part of the variance in BC concentrations. However, the overall predictive performance remains low, due to input uncertainty and lack of predictive variables that can properly capture the complex characteristics of local concentrations. We stress the importance of using an appropriate cross-validation scheme to estimate the predictive performance of the model. By using independent data for the validation and excluding those data also during variable selection in the model building procedure, overly optimistic performance estimates are avoided. Land use regression models are built based on opportunistic mobile measurements.No significant difference between different regression techniques.Distinction between cross-validation with and without a full rebuild of the model.Importance of an appropriate cross-validation scheme to estimate the performance.LUR models explain a significant part of the variance, but overall predictive performance is low.