MLP bilinear separation

In this paper, we present a thorough mathematical analysis of the use of neural networks to solve a specific classification problem consisting of a bilinear boundary. The network under consideration is a three-layered perceptron with two hidden neurons having the sigmoid serving as the activation function. The analysis of the hidden space created by the outputs of the hidden neurons will provide results on the network’s capacity to isolate two classes of data in a bilinear fashion, and the importance of the value of the sigmoid parameter is highlighted. We will obtain an explicit analytical function describing the boundary generated by the network, thus providing information on the effect each parameter has on the network’s behavior. Generalizations of the results are obtained with additional neurons, and a theorem concerned with analytical reproducibility of the boundary function is established.     

Study on the replication quality of micro-structures in the injection molding process with dynamical tool tempering systems

The injection molding of micro-structures is a promising mass-production method for a broad range of materials. However, the replication quality of these structures depends significantly on the heat flow during the filling stage. In this paper, the filling and heat transfer of v-groove and random structures below 5 μm is investigated with the help of an AFM (atomic force microscope) and thermo couples. A numerical model is developed to predict the filling of surface structures during the filling and packing stage. The model implies the use of simple fully developed flow models taking the power-law material model into account. This permits investigation into which ways several processing parameters affect the polymer flow in the surface structures. The mold wall temperature, which has significant effects on the polymer flow, is varied by using a variothermal mold temperature control system to validate the model proposed.     

Improving component selection and monitoring with controlled experimentation and automated measurements

Context: A number of approaches have been proposed for the general problem of software component evaluation and selection. Most approaches come from the field of Component-Based Software Development (CBSD), tackle the problem of Commercial-off-the-shelf component selection and use goal-oriented requirements modelling and multi-criteria decision making techniques. Evaluation of the suitability of components is carried out largely manually and partly relies on subjective judgement. However, in dynamic, distributed environments with high demands for transparent selection processes leading to trustworthy, auditable decisions, subjective judgements and vendor claims are not considered sufficient. Furthermore, continuous monitoring and re-evaluation of components after integration is sometimes needed.Objective: This paper describes how an evidence-based approach to component evaluation can improve repeatability and reproducibility of component selection under the following conditions: (1) Functional homogeneity of candidate components and (2) High number of components and selection problem instances.Method: Our evaluation and selection method and tool empirically evaluate candidate components in controlled experiments by applying automated measurements. By analysing the differences to system-development-oriented scenarios, the paper shows how the process of utility analysis can be tailored to fit the problem space, and describes a method geared towards automated evaluation in an empirical setting. We describe tool support and a framework for automated measurements.We further present a taxonomy of decision criteria for the described scenario and discuss the data collection means needed for each category of criteria.Results: To evaluate our approach, we discuss a series of case studies in the area of digital preservation. We analyse the criteria defined in these case studies, classify them according to the taxonomy, and discuss the quantitative coverage of automated measurements.Conclusion: The results of the analysis show that an automated measurement, evaluation and selection framework is necessary and feasible to ensure trusted and repeatable decisions.     

Source/Filter Model for Unsupervised Main Melody Extraction From Polyphonic Audio Signals

Extracting the main melody from a polyphonic music recording seems natural even to untrained human listeners. To a certain extent it is related to the concept of source separation, with the human ability of focusing on a specific source in order to extract relevant information. In this paper, we propose a new approach for the estimation and extraction of the main melody (and in particular the leading vocal part) from polyphonic audio signals. To that aim, we propose a new signal model where the leading vocal part is explicitly represented by a specific source/filter model. The proposed representation is investigated in the framework of two statistical models: a Gaussian Scaled Mixture Model (GSMM) and an extended Instantaneous Mixture Model (IMM). For both models, the estimation of the different parameters is done within a maximum-likelihood framework adapted from single-channel source separation techniques. The desired sequence of fundamental frequencies is then inferred from the estimated parameters. The results obtained in a recent evaluation campaign (MIREX08) show that the proposed approaches are very promising and reach state-of-the-art performances on all test sets.      

“Those found responsible have been sacked”: some observations on the usefulness of error

Erik Hollnagel’s body of work in the past three decades has molded much of the current research approach to system safety, particularly notions of “error”. Hollnagel regards “error” as a dead-end and avoids using the term. This position is consistent with Rasmussen’s claim that there is no scientifically stable category of human performance that can be described as “error”. While this systems view is undoubtedly correct, “error” persists. Organizations, especially formal business, political, and regulatory structures, use “error” as if it were a stable category of human performance. They apply the term to performances associated with undesired outcomes, tabulate occurrences of “error”, and justify control and sanctions through “error”. Although a compelling argument can be made for Hollnagel’s view, it is clear that notions of “error” are socially and organizationally productive. The persistence of “error” in management and regulatory circles reflects its value as a means for social control.     

A bounded distance metric for comparing tree structure

Comparing tree-structured data for structural similarity is a recurring theme and one on which much effort has been spent. Most approaches so far are grounded, implicitly or explicitly, in algorithmic information theory, being approximations to an information distance derived from Kolmogorov complexity. In this paper we propose a novel complexity metric, also grounded in information theory, but calculated via Shannon's entropy equations. This is used to formulate a directly and efficiently computable metric for the structural difference between unordered trees. The paper explains the derivation of the metric in terms of information theory, and proves the essential property that it is a distance metric. The property of boundedness means that the metric can be used in contexts such as clustering, where second-order comparisons are required. The distance metric property means that the metric can be used in the context of similarity search and metric spaces in general, allowing trees to be indexed and stored within this domain. We are not aware of any other tree similarity metric with these properties.