A service‐agnostic method for predicting service metrics in real time

We predict performance metrics of cloud services using statistical learning, whereby the behaviour of a system is learned from observations. Specifically, we collect device and network statistics from a cloud testbed and apply regression methods to predict, in real‐time, client‐side service metrics for video streaming and key‐value store services. Results from intensive evaluation on our testbed indicate that our method accurately predicts service metrics in real time (mean absolute error below 16% for video frame rate and read latency, for instance). Further, our method is service agnostic in the sense that it takes as input operating systems and network statistics instead of service‐specific metrics. We show that feature set reduction significantly improves the prediction accuracy in our case, while simultaneously reducing model computation time. We find that the prediction accuracy decreases when, instead of a single service, both services run on the same testbed simultaneously or when the network quality on the path between the server cluster and the client deteriorates. Finally, we discuss the design and implementation of a real‐time analytics engine, which processes streams of device statistics and service metrics from testbed sensors and produces model predictions through online learning.     

A Compact Two-dimensional Power Divider for a Dielectric Rod Antenna Array Based on Multimode Interference

In this work, multimode interference is investigated for the design of a two-dimensional fully dielectric power divider, well suited for the usage of dielectric waveguides. Most important, power division is achieved in a single device without the need of cascading multiple dividers. This allows to design a very compact and lightweight power divider, well applicable for dielectric rod antenna arrays. As a proof of concept for the used technique, a 16-way power divider with 4 × 4 output ports, made out of Rexolite, is realized, working in a frequency range between 90 and 105 GHz. For the S-parameter measurements, a special measurement setup, including a modular pin probe technique as well as radiation taper for waveguide termination, is proposed. The measurements are in good agreement with the simulations with a power split of ??15 dB for all output ports within the desired frequency range. This is equal to an additional insertion loss of 3 dB. To demonstrate the usability for antenna arrays, a fully dielectric rod antenna array is realized based on the proposed power divider. With this array, a gain of 22.5 dBi at 97.5 GHz was achieved.     

Learning programs by learning from failures

Machine Learning - We describe an inductive logic programming (ILP) approach called learning from failures. In this approach, an ILP system (the learner) decomposes the learning problem into three...     

Fehlertoleranz bei mechatronischen Systemen

Mechatronic systems are characterized by replacing mechanical, hydraulic or pneumatic components by electrical and electronic components and programmed controlled functions. They offer a relatively flexible design with extended functions. However, the failure behaviour of sensors, electronics, actuators and software is disadvantageous compared to the failure behaviour of pure mechanics. An introduction first considers the status of fault-tolerant systems with mobile applications and computers. Some basic fault-tolerant structures with hardware and analytical redundancy for sensors, actuators, electronics and software are described including their degradation behaviour. Then, some examples of realized, fault-tolerant systems are considered as, e.?g., fault-tolerant electrical drives, electrical and hydraulic actuators, and sensor systems for mechanical quantities. The applied fault-detection methods and the reconfiguration are described. Herewith is it shown how the function can be maintained with only small disturbances of the operation after appearance of faults.     

Modeling lightness perception—another point of view

In a recent note, Brill and Carter propose to revisit the centuries‐old disagreement over the relationship between stimuli and perceived lightness or brightness and the mathematical model that best represents it. Here, the answer is offered that the only way to resolve this matter is empirically, with controlled experiments of different kinds that establish statistically meaningful and replicated data for a given set of test conditions and methodology. Given the fact that surround lightness has a very significant effect on the results, including the crispening effect, a model can only be valid for a limited set of conditions, and the likely outcome is multiple models or models with multiple variables. © 2013 Wiley Periodicals, Inc. Col Res Appl, 39, 102–104, 2014     

Mechanical catalysis on the centimetre scale

Enzymes play important roles in catalysing biochemical transaction paths, acting as logical machines through the morphology of the processes. A key challenge in elucidating the nature of these systems, and for engineering manufacturing methods inspired by biochemical reactions, is to attain a comprehensive understanding of the stereochemical ground rules of enzymatic reactions. Here, we present a model of catalysis that can be performed magnetically by centimetre-sized passive floating units. The designed system, which is equipped with permanent magnets only, passively obeys the local causalities imposed by magnetic interactions, albeit it shows a spatial behaviour and an energy profile analogous to those of biochemical enzymes. In this process, the enzyme units trigger physical conformation changes of the target by levelling out the magnetic potential barrier (activation potential) to a funnel type and, thus, induce cascading conformation changes of the targeted substrate units reacting in parallel. The inhibitor units, conversely, suppress such changes by increasing the potential. Because the model is purely mechanical and established on a physics basis in the absence of turbulence, each performance can be explained by the morphology of the unit, extending the definition of catalysis to systems of alternative scales.     

Variability in estimation of suprathreshold small color differences

Intraobserver and interobserver variability in perceptual small color difference evaluation are known but are rarely analyzed phenomena. They raise the question if humans are equipped to make reliable and uniform estimates of this kind. Data from two recent well‐controlled studies show that intraobserver and, particularly, interobserver variability are surprisingly large. The use of a gray reference scale in place of a single gray reference pair has a small normative effect on the estimate. A large interobserver variability in the estimated magnitude of chromatic differences compared with achromatic reference pairs remains, with an average ratio between largest and smallest estimate of over 3:1 and ratios for individual observers and sample pairs ranging up to 28:1. No correlation was found between the reliability of judgment and the judged magnitude of difference: highly reliable observer groups are not more consistent. Relative reliability of estimate was found inversely related to magnitude of perceived difference. Consequences of these results for the development of improved color difference formulas are discussed. © 2009 Wiley Periodicals, Inc. Col Res Appl, 34, 367–374, 2009     

True Random Number Generator With a Metastability-Based Quality Control

We present a metastability-based true random number generator that achieves high entropy and passes NIST randomness tests. The generator grades the probability of randomness regardless of the output bit value by measuring the metastable resolution time. The system determines the original random noise level at the time of metastability and tunes itself to achieve a high probability of randomness. Dynamic control enables the system to respond to deterministic noise and a qualifier module grades the individual metastable events to produce a high-entropy random bit-stream. The grading module allows the user to trade off output bit-rate with the quality of the bit-stream. A fully integrated true random number generator was fabricated in a 0.13 mum bulk CMOS technology with an area of 0.145 mm².     

Comparison of unique hue stimuli determined by two different methods using Munsell color chips

Unique hue stimuli were determined by male and female observers using two different visual experimental procedures involving Munsell color chips of varying hue but identical chroma and value. The hypothesis was that unique hues can be more reliably established by explicit selection from a series of ordered stimuli than implicitly by hue scaling a series of stimuli in terms of neighboring UHs and this was statistically confirmed. The implicit selections based on long term memory of UHs appears to have been more challenging to observers since variability was increased by nearly 50% compared to when UHs were explicitly selected. The ranges of unique hues selected in the two methods were, however, comparable and no statistically significant difference was found between the results of females and males. The intra‐observer variability in picking a stimulus to represent a unique hue, for all observers and averaged for all hues, was approximately 12% of the mean spread of unique hues, confirming that the large inter‐observer variability is driven by differences in color vision and perhaps cognitive processes. © 2010 Wiley Periodicals, Inc. Col Res Appl, 2010