Recurrent neural networks and robust time series prediction

We propose a robust learning algorithm and apply it to recurrent neural networks. This algorithm is based on filtering outliers from the data and then estimating parameters from the filtered data. The filtering removes outliers from both the target function and the inputs of the neural network. The filtering is soft in that some outliers are neither completely rejected nor accepted. To show the need for robust recurrent networks, we compare the predictive ability of least squares estimated recurrent networks on synthetic data and on the Puget Power Electric Demand time series. These investigations result in a class of recurrent neural networks, NARMA(p,q), which show advantages over feedforward neural networks for time series with a moving average component. Conventional least squares methods of fitting NARMA(p,q) neural network models are shown to suffer a lack of robustness towards outliers. This sensitivity to outliers is demonstrated on both the synthetic and real data sets. Filtering the Puget Power Electric Demand time series is shown to automatically remove the outliers due to holidays. Neural networks trained on filtered data are then shown to give better predictions than neural networks trained on unfiltered time series.     

Sharpen your technique for needle-free injection

A clinical ladder identifies nursing competence, provides recognition and financial reward for bedside nursing practice, promotes responsibility, accountability, project and committee involvement, and enhances self-worth and positive morale. Because of the program's blend of skill levels, it develops a climate for clinical learning while promoting professional growth and confidence.     

Fitting Square Pegs into Round Holes

By nature, cryptographers tend to be extremely cautious and conservative. From their perspective, spending time getting people to use a new cryptographic algorithm is roughly on par with getting involved in a land war in Asia. In some cases, however, the difficulty involved in developing a new algorithm is worth the effort. This is the case with identity-based encryption (IBE), a new type of public-key algorithm     

The influence of the fabrication process on the buckling of thin-walled steel box sections

Steel box sections are usually fabricated from flat plates which are welded at the corners. The welding process can introduce residual stresses and geometric imperfections into the sections which can influence their strength. For some thin-walled sections, large periodic geometric imperfections have been observed in manufactured sections. Subsequent investigations have indicated that the imperfections are in fact buckling deformations i.e. the box section has buckled due to welding residual stresses prior to any application of external load. The welding procedure and the behaviour of the box sections under load has been modelled using a finite element analysis that accounts for both geometric and material non-linearities. Tests have been carried out on box sections with a range of width to thickness ratios for the plate elements. Modelling has been shown to give good correlation with the test results. The conditions for buckling to take place as a result of the welding process have been established. A design method has been proposed.     

On the Automatic Recovery of Style-Specific Architectural Relations in Software Systems

The cost of maintaining a software system over a long period of time far exceeds its initial development cost. Much of the maintenance cost is attributed to the time required by new developers to understand legacy systems. High-level structural information helps maintainers navigate through the numerous low-level components and relations present in the source code. Modularization tools can be used to produce subsystem decompositions from the source code but do not typically produce high-level architectural relations between the newly found subsystems. Controlling subsystem interactions is one important way in which the overall complexity of software maintenance can be reduced.We have developed a tool, called ARIS (Architecture Relation Inference System), that enables software engineers to define rules and relations for regulating subsystem interactions. These rules and relations are called Interconnection Styles and are definedusing a visual notation. The style definition is used by our tool to infer subsystem-level relations in designs being reverse engineered from source code.In this paper we describe our tool and its underlying techniques and algorithms. Using a case study, we describe how ARIS is used to reverse engineer high-level structural information from a real application.     

Anfangsvorgänge bei Lichtenbergschen Figuren

Zusammenfassung Aufnahmen von Lichtenbergschen Figuren mit kurzen Spannungsstößen zeigen, daß bisher bekanntgewordene Bilder solcher Entladungsfiguren bereits ein verhältnismäßig weit vorgeschrittenes Stadium darstellen. Es gelingt mittels kurzer Stöße erste lawinenartige Anfänge der Figuren festzuhalten und einiges über Vorgänge im Anfangsstadium herauszulesen. Die negative Figur beginnt lawinenförmig an bevorzugten Stellen der Kathode. Bei nebeneinander entstandenen Lawinen ist sehr gut die gegenseitige Abstoßung infolge gleicher Ladung zu beobachten. Die Entladung hat einen geschichteten Aufbau. Die positiven Figuren bilden sich später aus. Es sprechen jedoch bisher keinerlei Anzeichen für ein Hinwachsen der Fäden zur Anode. Die Kanäle verjüngen sich zur Anode hin und selbst bei kurzen Stoßzeiten endet kein einziger vor der Anode. Auch die Art der gegenseitigen Abstoßung der positiven Fäden spricht mehr für ein Vorwachsen von der Anode aus. Aus den Farbaufnahmen ergibt sich für die filmschwärzende Strahlung eine Wellenlänge unter 450 m.     

Adaptive user modelling in intelligent telephone and email assistants

With the burgeoning complexity and capabilities of modern information appliances and services, user modelling is becoming an increasingly important research area. Simple user profiles already personalise many software products and consumer goods such as digital TV recorders and mobile phones. A user model should be easy to initialise, and it must adapt in the light of interaction with the user. In many cases, a large amount of training data is needed to generate a user model, and adaptation is equivalent to retraining the system. This paper briefly outlines the user modelling problem and work done at BTexact on an intelligent personal assistant (IPA) which incorporates a user profile. We go on to describe FILUM, a more flexible method of user modelling, and show its application to the telephone assistant and email assistant components of the IPA, with tests to illustrate its usefulness.     

Study of a BGO calorimeter using electron and hadron beams from 1 to 50 GeV

A calorimeter of 25 bismuth germanate (BGO) crystals equipped with silicon photodiode readout has been tested at the CERN SPS in the energy range 1–50 GeV. The response for electrons has been shown to be linear in this energy range and the rms resolution obtained ( ) is approximately 1%, for E > 4 GeV. The electron/pion separation was found to be better than 1:500 in the energy range 1–20 GeV. Data on lateral and longitudinal shower development were compared with the results of a Monte Carlo simulation using the SLAC-EGS program and found to be in good agreement.