Comments on “Can backpropagation error surface not have localminima?”

in the above paper Yu (IEEE Trans. Neural Networks, vol.3, no.6, p.1019-21 (1992)) claims to prove that local minima do not exist in the error surface of backpropagation networks being trained on data with t distinct input patterns when the network is capable of exactly representing arbitrary mappings on t input patterns. The commenter points out that the proof presented is flawed, so that the resulting claims remain unproved. In reply, Yu points out that the undesired phenomenon that was sited can be avoided by simply imposing the arbitrary mapping capacity of the network on lemma 1 in the article.     

A practical application of probabilistic modelling in assessment of dietary exposure of fruit consumers to pesticide residues.

In 1996, studies on a range of organophosphate and carbamate pesticide residues in fruit that may be eaten as single items reported variability. The usual point estimate exposure model did not take account of the variation in residue levels between items or variation in consumption patterns of individual consumers. Using only the highest residue levels and consumption values for each of the multiple sources (different fruit) could lead to overestimates of residue intakes which would indicate higher than actual levels of risk. Probabilistic simulation was identified as a tool that could utilize all the available information from the variability studies and fruit consumption data collected from dietary surveys. The estimation of exposure of toddlers to carbaryl is shown as an example. The number of samples representing some combinations of fruit in the toddler dietary survey was particularly low and the validity of extrapolating from these was unknown. Therefore, consumption values were simulated using the data for frequency and amount eaten from the whole database. The data indicated that there were some weak positive associations between consumption levels of the different fruit. However, inclusion of correlated sampling in the model simulation was considered too conservative. The profiles of carbaryl residues in different retail batches differed. Therefore a model was constructed that differentiated between different residue profiles and sampled separate residue levels for each item assumed to be eaten. Two simpler models, both ignoring the effect of re-sampling from the same batch, were also used to estimate exposure. All three models were considered to give realistic views of the likely short-term intakes and the outputs were useful as an aid to decision-making in terms of necessary regulatory action.     

Implementing a Domain-Specific Language Using Stratego/XT: An Experience Paper

We describe the experience of implementing a Domain-Specific Language using transformation to a General Purpose Language. The domain of application is image processing and low-level computer vision. The transformation is accomplished using the Stratego/XT language transformation toolset. The implementation presented here is contrasted with the original implementation carried out many years ago using standard compiler implementation tools of the day. We highlight some of the unexpected advantages afforded to us, as language designers and implementers, by the source-to-source transformation technique. We also present some of the practical challenges faced in the implementation and show how these issues were addressed.     

Mixed-Mode Fracture of Hybrid Material Bonded Interfaces under Four-Point Bending

A combined analytical and experimental approach is presented to characterize mixed-mode fracture of hybrid material bonded interfaces under four-point bending load, and closed-form solutions of compliance and energy release rate (ERR) of the mixed-mode fracture specimens are provided. The transverse shear deformations in each sublayer of bimaterial bonded beams are included by modeling the specimen as individual Timoshenko beams, and the effect of interface crack-tip deformation on the compliance and ERR are taken into account by applying the interface deformable bilayer beam theory (flexible-joint model). The higher accuracy of the present analytical solutions for both the compliance and ERR of mixed-mode fracture specimens is manifested by comparing them with the solutions predicted by the conventional beam theory (CBT) and finite-element analysis (FEA). As an application example, the fracture of wood–fiber-reinforced plastic (FRP) bonded interface is experimentally evaluated by using mixed-mode fracture specimens [i.e., four-point asymmetric end-notched flexure (4-AENF) and four-point mixed-mode bending (4-MMB)], and the corresponding values of critical ERRs are obtained. Comparisons of the compliance rate change and the resulting critical ERR based on the CBT, the present theoretical model, and FEA demonstrate that the crack-tip deformation plays an important role in accurately characterizing the mixed-mode fracture toughness of hybrid material bonded interfaces under four-point bending load.     

A new model for the prediction of agricultural operator exposure during professional application of plant protection products in outdoor crops

A new predictive model for the estimation of agricultural operator exposure has been developed on the basis of new exposure data to improve the current agricultural operator exposure and risk assessment in the EU. The new operator exposure model represents current application techniques and practices in EU Member States (MS) and is applicable for national or zonal authorisation of plant protection products as well as for approval of active substances in plant protection products (PPP) supporting a stepwise risk assessment. 34 unpublished exposure studies conducted between 1994 and 2009 were evaluated for the new model. To ensure a very high quality of data the studies had to meet a set of quality criteria, e.g. GLP conformity or compliance with OECD guidance. Exposure data and supplementary information on the trials were used for a statistical analysis of exposure factors. The statistical analyses resulted in six validated models for typical outdoor scenarios of pesticide mixing/loading and application. As a major factor contributing to the exposure of operators, the amount of active substance used per day was identified. Other parameters such as formulation type, droplet size, presence of a cabin or density of the canopy were selected as factors for sub-scenarios. For two scenarios the corresponding datasets were too small to identify reliable exposure factors; instead the relevant percentiles of the exposure distribution were used. The whole project report on the development of the new model (including the underlying study data) and the corresponding exposure calculators will be published at the BfR website (http://www.bfr.bund.de) after confirmation of the model by EU MS.     

Occurrence and risks associated with chlormequat residues in a range of foodstuffs in the UK

Trends in the usage pattern of chlormequat (a plant growth regulator) on cereal crops in the UK over the past 6 years are presented. The figures have been supplemented by monitoring of chlormequat residues in food commodities through the UK surveillance programme, and for cereals, the carry-through into a number of processed cereal-based foods has been followed. A downward trend of chlormequat residue levels in pears sampled between 1997 and 2002 was observed. This decline reflects changes that were introduced in European and national regulations and which have proven to be extremely effective in reducing both the frequency of detection and levels of chlormequat residues. Both acute and chronic risk assessments were undertaken based on aggregate dietary exposure data. Even when the highest residues observed were used in the calculations, the assessments showed that both the short- and long-term intakes for all consumer groups would be unlikely to cause adverse health effects and were therefore not of any cause for concern.     

Nano-milling of pigment agglomerates using a wet stirred media mill: Elucidation of the kinetics and breakage mechanisms

The present study concerns the production of pigment nanoparticles in a wet-batch stirred media mill with polymeric media. The breakage kinetics and mechanisms were investigated using size-discrete population balance models (PBMs). The temporal variation of the particle size distribution was measured via dynamic light scattering. Considering the G-H model, a time-invariant PBM, and a time-variant PBM, the specific breakage rate parameters and breakage distribution parameters were identified. It is found that the breakage rate is not first-order and that a delay time exists for the breakage of nanoparticles. The time-variant PBM captures all these features and suggests a transition from deagglomeration of agglomerates to the breakage of primary particles. The analysis of the breakage distribution parameters suggests splitting as the dominant mechanism as opposed to attrition or massive fracture.