PAC-Bayesian Compression Bounds on the Prediction Error of Learning Algorithms for Classification

We consider bounds on the prediction error of classification algorithms based on sample compression. We refine the notion of a compression scheme to distinguish permutation and repetition invariant and non-permutation and repetition invariant compression schemes leading to different prediction error bounds. Also, we extend known results on compression to the case of non-zero empirical risk.We provide bounds on the prediction error of classifiers returned by mistake-driven online learning algorithms by interpreting mistake bounds as bounds on the size of the respective compression scheme of the algorithm. This leads to a bound on the prediction error of perceptron solutions that depends on the margin a support vector machine would achieve on the same training sample.Furthermore, using the property of compression we derive bounds on the average prediction error of kernel classifiers in the PAC-Bayesian framework. These bounds assume a prior measure over the expansion coefficients in the data-dependent kernel expansion and bound the average prediction error uniformly over subsets of the space of expansion coefficients.Editor Shai Ben-David     

Eine Dienstgüte unterstützende Webservice-Architektur für flexible Geschäftsprozesse

Web services as a technology to enable distributed business processes gain in importance. However, the support of Quality of Service (QoS) is crucial in this context. Thus, we present an integrated web service architecture with comprehensive QoS support. The architecture we introduce in this paper supports the assessment of web services to assure that only web services will be used in critical business processes that satisfy the requirements defined by the user. The selection and execution of a certain web service depends on its QoS-properties that must be guaranteed in Service Level Agreements (SLAs). The compliance with SLAs is monitored by the system as well. Furthermore we introduce a prototypical implementation of our web service architecture.     

Arguments and Misunderstandings: Fuzzy Unification for Negotiating Agents

In this paper, we develop the notion of fuzzy unification and incorporate it into a novel fuzzy argumentation framework for extended logic programming. We make the following contributions: The argumentation framework is defined by a declarative bottom-up fixpoint semantics and an equivalent goal-directed top-down proofprocedure for extended logic programming. Our framework allows one to represent positive and explicitly negative knowledge, as well as uncertainty. Both concepts are used in agent communication languages such as KQML and FIPA ACL. One source of uncertainty in open systems stems from mismatches in parameter and predicate names and missing parameters. To this end, we conservatively extend classical unification and develop fuzzy unification based on normalised edit distance over trees.     

Dynamic behavior of a liquid/liquid interface at an oscillating wall

The dynamic behavior of the meniscus between two liquids has been investigated experimentally using mercury and silicon oil and theoretically by solution of the equations of motion applying the marker and cell (MAC) method. The liquids were contained in a plexiglass tube which could be moved relative to the liquids with constant and superposed sinusoidal velocities. Special attention was focused on the movement of the three-phase contact line (TPL) at the wall. Depending on the parameters of the motion, the TPL may be drawn above the flat part of the oil/mercury meniscus so that the normally convex contour of the meniscus disappears and a concave contour is formed. Good agreement was found between the measured and computed shapes of the meniscus.     

Elektrochemische und werkstoffphysikalische Untersuchungen von Laserstrahl-umgeschmolzenen Eisenwerkstoffen

Electrochemical and material-physical investigations of laser melted iron materials The surfaces of pure iron, St 38 and KT 45 samples were melted by a CO₂-cw laser. Electrochemical and material-physical properties of the generated surface layers were investigated. There exist significant differences between untreated and laser melted samples, whose possible causes are discussed.     

Determination of the threshold of gravity for inducing kinetosis in fish: A drop-tower experiment

It has been repeatedly shown earlier that some fish of a given batch reveal motion sickness (a kinetosis) at the transition from 1 g to microgravity. In the course of parabolic aircraft flight experiments, it has been demonstrated that kinetosis susceptibility is correlated with asymmetric inner ear otoliths (i.e., differently weighed statoliths on the right and the left side of the head) or with genetically predispositioned malformed cells within the sensory epithelia of the inner ear. Hitherto, the threshold of gravity perception for inducing kinetotic behaviour as well as the relative importance of asymmetric otoliths versus malformed epithelia for kinetosis susceptibility has yet not been determined. The following experiment using the ZARM drop-tower facility in Bremen, Germany, is proposed to be carried out in order to answer the aforementioned questions. Larval cichlid fish (Oreochromis mossambicus) will be kept in a camcorder-equipped centrifuge during the microgravity phases of the drops and thus receive various gravity environments ranging from 0.1 to 0.9 g. Videographed controls will be housed outside of the centrifuge receiving 0 g. Based on the videorecordings, animals will be grouped into kinetotically and normally swimming samples. Subsequently, otoliths will be dissected and their size and asymmetry will be measured. Further investigations will focus on the numerical quantification of inner ear supporting and sensory cells as well as on the quantification of inner ear carbonic anhydrase reactivity. A correlation between (1) the results to be obtained concerning the g-loads inducing kinetosis and (2) the corresponding otolith asymmetry/morphology of sensory epithelia/carbonic anhydrase reactivity will further contribute to the understanding of the origin of kinetosis susceptibility. Besides an outline of the proposed principal experiments, the present study reports on a first series of drop-tower tests which were undertaken to elucidate the feasibility of the proposal (especially concerning the question, if some 4.7 s of microgravity are sufficient to induce kinetotic behaviour in larval fish).     

Subroutines for some plasma surface interaction processes: physical sputtering, chemical erosion, radiation enhanced sublimation, backscattering and thermal evaporation

A suite of FORTRAN subroutines/functions to generate data using empirical formulas for physical sputtering of mono-atomic targets for any elemental incident ion (atom), chemical erosion of graphite, Radiation Enhanced Sublimation (RES) of graphite, the number and energy backscattering coefficients for any elemental incident ion (atom) on a compound target and thermal evaporation of graphite is presented. Since chemical erosion, RES and thermal evaporation depend on the surface temperature of graphite, a subroutine implementing the 1-D heat diffusion equation to determine the temperature of any plasma-facing graphite surface is implemented. As an example to illustrate the use of these subroutines/functions, a simple model for the erosion of a plasma-facing surface, consisting of a simple collisionless sheath model, a 1-dimensional steady state heat diffusion model and 0-dimensional steady state particle balance at the target is developed and a sample listing of the program is presented.

Program summary

Title of program: Plasma Surface Interaction Codes (PSIC)Catalog identifier: ADTRProgram summary URL:http://cpc.cs.qub.ac.uk/summaries/ADTRProgram obtainable from: CPC Program Library, Queen's University of Belfast, N. IrelandComputers for which the program has been designed for and others on which it has been tested: Program is designed for any computer with a FORTRAN-77 compiler. It has been tested on a Linux PC with g77, Absoft f77, f90, f95, Fujitsu f95, Lahey f95Operating systems under which the program has been tested: Linux, UNIXProgramming language used: FORTRAN-77Memory required to execute with typical data: Negligible (executable is 178 861 bytes; add to it the memory used by the Fortran library linked to)No. of bits in a word: 16No. of processors used: 1Has the code been vectorized or parallelized?: NoNo. of bytes in distributed program, including test data, etc.: 11 619No. of lines in distributed program, including test data, etc.: 1015Distribution format: tar gzip fileNature of physical problems: 0-D plasma surface interaction model used to demonstrate calls to subroutines for physical sputtering yield, chemical erosion yield of graphite, RES of graphite, backscattering coefficients, thermal evaporation of graphite and 1-D heat diffusion in graphite.Method of solution: Semi-empirical formulas are used for the PSIs and analytical formulas are used for thermal evaporation and 1-D heat diffusion in graphite.Restrictions on the complexity of the problem: All the PSIs (except for the backscattering coefficients) are valid only for a monoatomic target. The heat diffusion solution is not valid for non-uniform plasmas and for targets for which the heat diffusion is not isotropic.Typical running time: Example program takes much less than 0.01 secondUniversal features of the program: None