MLP in layer-wise form with applications to weight decay

A simple and general calculus for the sensitivity analysis of a feedforward MLP network in a layer-wise form is presented. Based on the local optimality conditions, some consequences for the least-means-squares learning problem are stated and further discussed. Numerical experiments with formulation and comparison of different weight decay techniques are included.     

EEG and MEG brain-computer interface for tetraplegic patients.

We characterized features of magnetoencephalographic (MEG) and electroencephalographic (EEG) signals generated in the sensorimotor cortex of three tetraplegics attempting index finger movements. Single MEG and EEG trials were classified offline into two classes using two different classifiers, a batch trained classifier and a dynamic classifier. Classification accuracies obtained with dynamic classifier were better, at 75%, 89%, and 91% in different subjects, when features were in the 0.5-3.0-Hz frequency band. Classification accuracies of EEG and MEG did not differ.     

Compatibilization of polypropylene/poly(n-butylacrylate) blend: Functionalization of the poly(n-butylacrylate) phase by copolymerization

Compatibilization of a polypropylene (PP)/polybutylacrylate (PBuA) blend was studied, with the aim of achieving better adhesion at the interphase through modification of both the PP and PBuA phases. The compatibilization involved two separate stages: First, a small amount (2 and 5 mol %) of functional monomer copolymerizable with and soluble in BuA was added to the BuA initiator solution. Then, this solution was impregnated into PP pellets and polymerized inside the pellets by free-radical polymerization in a water suspension. The resulting blend was a thermoplastic elastomer consisting of PP as the matrix and functionalized PBuA as the partly crosslinked dispersed rubbery phase. The functionalities of the monomers were epoxy, oxazoline, hydroxyl, secondary amine, and carboxyl. In the second stage, two commercial graft copolymers of PP (PP grafted with either acrylic acid or maleic acid anhydride) were blended at the melt stage with the PP/functionalized PBuA blend. Here, the compatibilizing reactions took place between the functionalized components of the blend. The compatilization reactions were detected by FTIR analysis and by changes in mechanical or thermomechanical behavior of the blends. Morphology studies were carried out. As a result, the tensile strength, tensile modulus, elongation at break, and tear strength of the final product were improved by about 15, 20, 160, and 50%, respectively, compared with the unfunctionalized blend. The hardness of the material remained unchanged in the compatibilization. © 1996 John Wiley & Sons, Inc.     

Residual stresses in HVOF-sprayed ceramic coatings

In this paper, the residual stress state of thermally sprayed ceramic coatings was examined by combining different experimental and analytical techniques, in order to provide a thorough characterisation of through-thickness stress profiles and a cross-verification of results. HVOF-sprayed ceramics, manufactured using commercial and nanostructured Al₂O₃ powders and commercial Cr₂O₃ powders, and atmospheric plasma-sprayed (APS) ceramics, manufactured using commercial Al₂O₃ and Cr₂O₃ powders, were investigated.The near-surface stress was measured by X-ray diffraction. The through-thickness profile and the intrinsic quenching stress were analytically computed by the Tsui-Clyne iterative model, using the X-ray measurement result as input, and results were validated by the substrate chemical removal method. Further verification was achieved by applying the in-situ curvature technique to the deposition of HVOF-sprayed Al₂O₃ coating.HVOF-sprayed Al₂O₃ coatings deposited using both conventional and nanostructured powders feature a similar, almost equibiaxial tensile stress on the top surface (116.5 MPa and 136.5 MPa, respectively) and a moderate through-thickness gradient (about 12 MPa and 20 MPa, respectively). Their intrinsic quenching stresses were analytically estimated to be 184 MPa and 205 MPa, respectively. APS Al₂O₃ possesses higher top surface stress (220 MPa) and quenching stress (311 MPa). However, it shows a less pronounced stress gradient (≈ 3 MPa) than HVOF-sprayed Al₂O₃-based coatings, because cracks, pores and weak lamella boundaries in the APS coating can accommodate the deformations induced by the bending moments arising both during coating deposition and during cooling.The model-derived quenching stress of the conventional HVOF Al₂O₃ coating was validated by the in-situ curvature measurement technique.Cr₂O₃-based coatings are significantly different. They display a lower residual stress in the near-surface region: 20 MPa in the APS coating, 27.5 MPa in the HVOF one. The HVOF coating also exhibits a very large stress gradient of ≈ 77 MPa. Machining and sliding processes (like polishing and dry sliding tribological testing) change their surface residual stresses to compressive ones.     

Recent advances in Apertium,a free/open-source rule-based machine translation platform for low-resource languages

This paper presents an overview of Apertium, a free and open-source rule-based machine translation platform. Translation in Apertium happens through a pipeline of modular tools, and the platform continues to be improved as more language pairs are added. Several advances have been implemented since the last publication, including some new optional modules: a module that allows rules to process recursive structures at the structural transfer stage, a module that deals with contiguous and discontiguous multi-word expressions, and a module that resolves anaphora to aid translation. Also highlighted is the hybridisation of Apertium through statistical modules that augment the pipeline, and statistical methods that augment existing modules. This includes morphological disambiguation, weighted structural transfer, and lexical selection modules that learn from limited data. The paper also discusses how a platform like Apertium can be a critical part of access to language technology for so-called low-resource languages, which might be ignored or deemed unapproachable by popular corpus-based translation technologies. Finally, the paper presents some of the released and unreleased language pairs, concluding with a brief look at some supplementary Apertium tools that prove valuable to users as well as language developers. All Apertium-related code, including language data, is free/open-source and available at https://github.com/apertium.

     

Common cause failure analysis of cyber–physical systems situated in constructed environments

While cyber–physical system sciences are developing methods for studying reliability that span domains such as mechanics, electronics and control, there remains a lack of methods for investigating the impact of the environment on the system. External conditions such as flooding, fire or toxic gas may damage equipment and failing to foresee such possibilities will result in invalid worst-case estimates of the safety and reliability of the system. Even if single component failures are anticipated, abnormal environmental conditions may result in common cause failures that cripple the system. This paper proposes a framework for modeling interactions between a cyber–physical system and its environment. The framework is limited to environments consisting of spaces with clear physical boundaries, such as power plants, buildings, mines and urban underground infrastructures. The purpose of the framework is to support simulation-based risk analysis of an initiating event such as an equipment failure or flooding. The functional failure identification and propagation (FFIP) framework is extended for this purpose, so that the simulation is able to detect component failures arising from abnormal environmental conditions and vice versa: Flooding could be caused by a failure in a pipe or valve component. As abnormal flow states propagate through the system and its environment, the goal of the simulation is to identify the system-wide cumulative effect of the initiating event and any related common cause failure scenario. FFIP determines this effect in terms of degradation or loss of the functionality of the system. The method is demonstrated with a nuclear reactor’s redundant coolant supply system.     

The performance and efficacy of a two‐person operated portable PIT‐antenna for monitoring spatial distribution of stream fish populations

Passive Integrated Transponder (PIT) technology has enabled the tracking of individual fish by using data‐logging stations and one‐person operated portable tracking units. Whereas data‐logging stations are fixed at a certain location, the portable units have traditionally been used in studies that require tracking operations at small spatial scales (<1000 m²) due to the time investment necessary to fully scan the stream channel. We developed a two‐person operated portable antenna that is stretched between the operators and the area in between is scanned for PIT‐tags. The antenna was five metres wide but can be customized to fit stream‐specific needs. The antenna can be used with existing Texas Instruments Series 2000 tracking units, and the detection distance ranged between 46 and 61 cm when using 23‐ or 32‐mm PIT‐tags, respectively (tag held parallel to the plane of the open coil inductor loop). To assess antenna performance in field trial, we compared the efficacy (% tags found) and time‐efficiency (time used to track a study site) between one‐ and two‐person antennae in three separate stream sites. The new antenna type proved to be very efficient (95.2–100%) for locating tags in all trials. However, the new antenna type might not perform adequately in areas with high structural complexity (e.g. logjams, overhanging vegetation) and a follow‐up with one‐person operated antenna is recommended in such complex habitats. The time consumption was always lower (10–53%) with the new antenna and significant time savings can be assumed especially in areas with a low density of PIT‐tagged individuals. Further, we have successfully used the new antenna type to track juvenile Atlantic salmon both in winter conditions (ice‐free) and warmer water (14°C) without causing fright responses. The new antenna type makes the use of PIT‐technology feasible not only at the micro/mesohabitat scale but also at the segment/reach scale and can thus be used to monitor behavioural responses of fish at the population level in wadable streams. Copyright © 2007 John Wiley & Sons, Ltd.     

Mine Water as a Resource: Selective Removal and Recovery of Trace Antimony from Mine-Impacted Water

The objective of this study was to evaluate the feasibility of combined modular processes to selectively remove Sb from mine-impacted waters in an Arctic environment in order to fulfil local environmental criteria for discharged waters. Novel ion exchange, selective extraction and ultrafiltration, electrocoagulation, and dissolved air flotation technologies were investigated, individually or in combination, from the laboratory- to pilot-demonstration scale. Laboratory-scale testing using Fe₂(SO₄)₃ precipitation, ion exchange resin, selective ion extraction and ultrafiltration, and electrocoagulation with or without subsequent dissolved air flotation indicated that any of the methods are potentially applicable to Sb removal from mine water. The observed differences between Sb and As removal efficiency by ion exchange resin illustrated the need for Sb-specific removal and recovery technologies. Techno-economic analyses showed that treatment of mine water using electrocoagulation-dissolved air flotation yields the lowest comparative life-cycle cost of examined technologies. Results demonstrated increased Sb attenuation efficiency using either electrocoagulation-dissolved air flotation or selective extraction and ultrafiltration, even when treating only 50% of the mine-impacted water, compared with conventional Fe₂(SO₄)₃ precipitation from mine water. Additional investigation is necessary to characterize the long-term stability of the mineral phases in Sb-containing solid residues and to inform selection of Sb recovery methods and utilisation or final disposal options for the residual materials.

     

Semi-supervised context adaptation: case study of audience excitement recognition

To recognise just the same human reaction (for example, a strong excitement) in different contexts, customary behaviours in these contexts have to be taken into account; e.g. a happy sport audience may be cheering for long time, while a happy theatrical audience may produce only short bursts of laughter in order to not interrupt the performance. Tailoring recognition algorithms to contexts can be achieved by building either a context-specific or a generic system. The former is individually trained for each context to recognise sets of characteristic responses, whereas the latter??in contrast to the context-specific one??adapts to the context via significantly more lightweight modification of parameters. This paper follows the latter way and proposes a simple modification of a hidden Markov model (HMM) classifier that enables end users to adapt the generic system to a context or a personal perception of an annotator by labelling a fairly small number of data samples of each context. For better adaptability to the limited number of the user??s annotations, the proposed semi-supervised HMM classifier employs the maximum posterior marginal, rather than the more conventional maximum a posteriori decision rule. The proposed user- and context-adaptable semi-supervised HMM classifier was tested on recognising excitement of a show audience in three contexts (a concert hall, a circus, and a sport event), differing in how the excitement is expressed. In our experiments the proposed classifier recognised reactions of a non-neutral audience with 10% higher accuracy than the conventional HMM and support vector machine based classifiers.