A benchmark-driven modelling approach for evaluating deployment choices on a multi-core architecture

The complexity of current and emerging architectures provides users with options about how best to use the available resources, but makes predicting performance challenging. In this work a benchmark-driven model is developed for a simple shallow water code on a Cray XE6 system, to explore how deployment choices such as domain decomposition and core affinity affect performance. The resource sharing present in modern multi-core architectures adds various levels of heterogeneity to the system. Shared resources often includes cache, memory, network controllers and in some cases floating point units (as in the AMD Bulldozer), which mean that the access time depends on the mapping of application tasks, and the core's location within the system. Heterogeneity further increases with the use of hardware-accelerators such as GPUs and the Intel Xeon Phi, where many specialist cores are attached to general-purpose cores. This trend for shared resources and non-uniform cores is expected to continue into the exascale era. The complexity of these systems means that various runtime scenarios are possible, and it has been found that under-populating nodes, altering the domain decomposition and non-standard task to core mappings can dramatically alter performance. To find this out, however, is often a process of trial and error. To better inform this process, a performance model was developed for a simple regular grid-based kernel code, shallow. The code comprises two distinct types of work, loop-based array updates and nearest-neighbour halo-exchanges. Separate performance models were developed for each part, both based on a similar methodology. Application specific benchmarks were run to measure performance for different problem sizes under different execution scenarios. These results were then fed into a performance model that derives resource usage for a given deployment scenario, with interpolation between results as necessary.

Another prefiltering algorithm for adaptive IIR filtering

An algorithm for adaptive IIR filtering that uses prefiltering structure in direct form is presented. This structure has an estimation error that is a linear function of the coefficients. This property greatly simplifies the derivation of gradient-based algorithms. Computer simulations show that the proposed structure improves convergence speed.

Constrained least square pre-distortion scheme for multiuser ultra-wideband

The study proposes a constrained least square (CLS) pre-distortion scheme for multiple-input single-output (MISO) multiple access ultra-wideband (UWB) systems. In such a scheme, a simple objective function is defined, which can be efficiently solved by a gradient-based algorithm. For the performance evaluation, scenarios CM1 and CM3 of the IEEE 802.15.3a channel model are considered. Results show that the CLS algorithm has a fast convergence and a good trade-off between intersymbol interference (ISI) and multiple access interference (MAI) reduction and signal-to-noise ratio (SNR) preservation, performing better than time-reversal (TR) pre-distortion.

Nonlinear Control Strategies for Cooperative Control of Multi-Robot Systems

This thesis deals with distributed control strategies for cooperative control of multi-robot systems. Specifically, distributed coordination strategies are presented for groups of mobile robots. The formation control problem is initially solved exploiting artificial potential fields. The purpose of the presented formation control algorithm is to drive a group of mobile robots to create a completely arbitrarily shaped formation. Robots are initially controlled to create a regular polygon formation. A bijective coordinate transformation is then exploited to extend the scope of this strategy, to obtain arbitrarily shaped formations. For this purpose, artificial potential fields are specifically designed, and robots are driven to follow their negative gradient. Artificial potential fields are then subsequently exploited to solve the coordinated path tracking problem, thus making the robots autonomously spread along predefined paths, and move along them in a coordinated way. Formation control problem is then solved exploiting a consensus based approach. Specifically, weighted graphs are used both to define the desired formation, and to implement collision avoidance. As expected for consensus based algorithms, this control strategy is experimentally shown to be robust to the presence of communication delays. The global connectivity maintenance issue is then considered. Specifically, an estimation procedure is introduced to allow each agent to compute its own estimate of the algebraic connectivity of the communication graph, in a distributed manner. This estimate is then exploited to develop a gradient based control strategy that ensures that the communication graph remains connected, as the system evolves. The proposed control strategy is developed initially for single-integrator kinematic agents, and is then extended to Lagrangian dynamical systems.

Adiabatic and local approximations for the kohn-sham potential in the hubbard model

We obtain the exact time-dependent Kohn-Sham potentials Vks for 1D Hubbard chains, driven by a d.c. external field, using the time-dependent electron density and current density obtained from exact many-body time-evolution. The exact Vxc is compared to the adiabatically-exact Vad-xc and the “instantaneous ground state” Vigs-xc. The effectiveness of these two approximations is analyzed. Approximations for the exchange-correlation potential Vxc and its gradient, based on the local density and on the local current density, are also considered and both physical quantities are observed to be far outside the reach of any possible local approximation. Insight into the respective roles of ground-state and excited-state correlation in the time-dependent system, as reflected in the potentials, is provided by the pair correlation function.

Dynamic modeling of active regeneration in catalyzed and non-catalyzed diesel particulate filters

Particulate matter (PM) emissions standards set by the US Environmental Protection Agency (EPA) have become increasingly stringent over the years. The EPA regulation for PM in heavy duty diesel engines has been reduced to 0.01 g/bhp-hr for the year 2010. Heavy duty diesel engines make use of an aftertreatment filtration device, the Diesel Particulate Filter (DPF). DPFs are highly efficient in filtering PM (known as soot) and are an integral part of 2010 heavy duty diesel aftertreatment system. PM is accumulated in the DPF as the exhaust gas flows through it. This PM needs to be removed by oxidation periodically for the efficient functioning of the filter. This oxidation process is also known as regeneration. There are 2 types of regeneration processes, namely active regeneration (oxidation of PM by external means) and passive oxidation (oxidation of PM by internal means). Active regeneration occurs typically in high temperature regions, about 500 - 600 °C, which is much higher than normal diesel exhaust temperatures. Thus, the exhaust temperature has to be raised with the help of external devices like a Diesel Oxidation Catalyst (DOC) or a fuel burner. The O2 oxidizes PM producing CO2 as oxidation product. In passive oxidation, one way of regeneration is by the use of NO2. NO2 oxidizes the PM producing NO and CO2 as oxidation products. The passive oxidation process occurs at lower temperatures (200 - 400 °C) in comparison to the active regeneration temperatures. Generally, DPF substrate walls are washcoated with catalyst material to speed up the rate of PM oxidation. The catalyst washcoat is observed to increase the rate of PM oxidation. The goal of this research is to develop a simple mathematical model to simulate the PM depletion during the active regeneration process in a DPF (catalyzed and non-catalyzed). A simple, zero-dimensional kinetic model was developed in MATLAB. Experimental data required for calibration was obtained by active regeneration experiments performed on PM loaded mini DPFs in an automated flow reactor. The DPFs were loaded with PM from the exhaust of a commercial heavy duty diesel engine. The model was calibrated to the data obtained from active regeneration experiments. Numerical gradient based optimization techniques were used to estimate the kinetic parameters of the model.

Individual and spousal education, mortality and life expectancy in Switzerland: a national cohort study.

BACKGROUND Household measures of socioeconomic position may better account for the shared nature of material resources, lifestyle, and social position of cohabiting persons, but household measures of education are rarely used. We aimed to evaluate the association of combined educational attainment of married couples on mortality and life expectancy in Switzerland. METHODS The study included 3 496 163 ever-married persons aged ≥30 years. The 2000 census was linked to mortality records through 2008. Mortality by combined educational attainment was assessed by gender-age-specific HRs, with 95% CIs from adjusted models, life expectancy was derived using abridged life tables. RESULTS Having a less educated partner was associated with increased mortality. For example, the HR comparing men aged 50-64 years with tertiary education married to women with tertiary education to men with compulsory education married to women with compulsory education was 2.05 (1.92-2.18). The estimated remaining life expectancy in tertiary educated men aged 30 years married to women with tertiary education was 4.6 years longer than in men with compulsory education married to women with compulsory education. The gradient based on individual education was less steep: the HR comparing men aged 50-64 years with tertiary education with men with compulsory education was 1.74 (1.67-1.81). CONCLUSIONS Using individual educational attainment of married persons is common in epidemiological research, but may underestimate the combined effect of education on mortality and life expectancy. These findings are relevant to epidemiologic studies examining socio-demographic characteristics or aiming to adjust results for these characteristics.

Decoding Delay Minimization in Inter-Session Network Coding

Intra-session network coding has been shown to offer significant gains in terms of achievable throughput and delay in settings where one source multicasts data to several clients. In this paper, we consider a more general scenario where multiple sources transmit data to sets of clients over a wireline overlay network. We propose a novel framework for efficient rate allocation in networks where intermediate network nodes have the opportunity to combine packets from different sources using randomized network coding. We formulate the problem as the minimization of the average decoding delay in the client population and solve it with a gradient-based stochastic algorithm. Our optimized inter-session network coding solution is evaluated in different network topologies and is compared with basic intra-session network coding solutions. Our results show the benefits of proper coding decisions and effective rate allocation for lowering the decoding delay when the network is used by concurrent multicast sessions.

The 1987 debris flows in Switzerland: documentation and analysis

A great number of debris flows occurred during the flood catastrophes of the summer of 1987 in the Swiss Alps. Aerial photography, field investigations and eyewitness accounts documented and analysed the events. As an example of the reconstructed major events, the large debris flow in the Varuna valley involved an estimated peak discharge between 400 and 800 m3/s and an event magnitude of 200,000 m3. Several single pulses were observed; the duration of each of them appeared to be not more than a few minutes. Apart from incision into weak bedrock, the maximum erosion depth seemed to depend on the channel gradient. Based on approximately 600 events, typical starting zones and rainfall conditions are discussed with regard to the triggering conditions. Existing and new empirical formulae are proposed to estimate the most important flow parameters. These values are compared to debris flow data from Canada and Japan.

Differentiating the Multipoint Expected Improvement for Optimal Batch Design

This work deals with parallel optimization of expensive objective functions which are modelled as sample realizations of Gaussian processes. The study is formalized as a Bayesian optimization problem, or continuous multi-armed bandit problem, where a batch of q > 0 arms is pulled in parallel at each iteration. Several algorithms have been developed for choosing batches by trading off exploitation and exploration. As of today, the maximum Expected Improvement (EI) and Upper Confidence Bound (UCB) selection rules appear as the most prominent approaches for batch selection. Here, we build upon recent work on the multipoint Expected Improvement criterion, for which an analytic expansion relying on Tallis’ formula was recently established. The computational burden of this selection rule being still an issue in application, we derive a closed-form expression for the gradient of the multipoint Expected Improvement, which aims at facilitating its maximization using gradient-based ascent algorithms. Substantial computational savings are shown in application. In addition, our algorithms are tested numerically and compared to state-of-the-art UCB-based batchsequential algorithms. Combining starting designs relying on UCB with gradient-based EI local optimization finally appears as a sound option for batch design in distributed Gaussian Process optimization.

Development of Shape-Optimization Tools for the Aerodynamic Design of Turbomachinery Blades

The aim of this work is to develop an automated tool for the optimization of turbomachinery blades founded on an evolutionary strategy. This optimization scheme will serve to deal with supersonic blades cascades for application to Organic Rankine Cycle (ORC) turbines. The blade geometry is defined using parameterization techniques based on B-Splines curves, that allow to have a local control of the shape. The location in space of the control points of the B-Spline curve define the design variables of the optimization problem. In the present work, the performance of the blade shape is assessed by means of fully-turbulent flow simulations performed with a CFD package, in which a look-up table method is applied to ensure an accurate thermodynamic treatment. The solver is set along with the optimization tool to determine the optimal shape of the blade. As only blade-to-blade effects are of interest in this study, quasi-3D calculations are performed, and a single-objective evolutionary strategy is applied to the optimization. As a result, a non-intrusive tool, with no need for gradients definition, is developed. The computational cost is reduced by the use of surrogate models. A Gaussian interpolation scheme (Kriging model) is applied for the estimated n-dimensional function, and a surrogate-based local optimization strategy is proved to yield an accurate way for optimization. In particular, the present optimization scheme has been applied to the re-design of a supersonic stator cascade of an axial-flow turbine. In this design exercise very strong shock waves are generated in the rear blade suction side and shock-boundary layer interaction mechanisms occur. A significant efficiency improvement as a consequence of a more uniform flow at the blade outlet section of the stator is achieved. This is also expected to provide beneficial effects on the design of a subsequent downstream rotor. The method provides an improvement to gradient-based methods and an optimized blade geometry is easily achieved using the genetic algorithm.

Hybrid convex combinations for IIR system identification.

The low complexity of IIR adaptive filters (AFs) is specially appealing to realtime applications but some drawbacks have been preventing their widespread use so far. For gradient based IIR AFs, adverse operational conditions cause convergence problems in system identification scenarios: underdamped and clustered poles, undermodelling or non-white input signals lead to error surfaces where the adaptation nearly stops on large plateaus or get stuck at sub-optimal local minima that can not be identified as such a priori. Furthermore, the non-stationarity in the input regressor brought by the filter recursivity and the approximations made by the update rules of the stochastic gradient algorithms constrain the learning step size to small values, causing slow convergence. In this work, we propose IIR performance enhancement strategies based on hybrid combinations of AFs that achieve higher convergence rates than ordinary IIR AFs while keeping the stability.

Non-linear analysis of the thermo-electro-mechanical behaviour of shear deformable FGM plates with piezoelectric actuators

This paper investigates the non-linear bending behaviour of functionally graded plates that are bonded with piezoelectric actuator layers and subjected to transverse loads and a temperature gradient based on Reddy's higher-order shear deformation plate theory. The von Karman-type geometric non-linearity, piezoelectric and thermal effects are included in mathematical formulations. The temperature change is due to a steady-state heat conduction through the plate thickness. The material properties are assumed to be graded in the thickness direction according to a power-law distribution in terms of the volume fractions of the constituents. The plate is clamped at two opposite edges, while the remaining edges can be free, simply supported or clamped. Differential quadrature approximation in the X-axis is employed to convert the partial differential governing equations and the associated boundary conditions into a set of ordinary differential equations. By choosing the appropriate functions as the displacement and stress functions on each nodal line and then applying the Galerkin procedure, a system of non-linear algebraic equations is obtained, from which the non-linear bending response of the plate is determined through a Picard iteration scheme. Numerical results for zirconia/aluminium rectangular plates are given in dimensionless graphical form. The effects of the applied actuator voltage, the volume fraction exponent, the temperature gradient, as well as the characteristics of the boundary conditions are also studied in detail. Copyright (C) 2004 John Wiley Sons, Ltd.

Efficient accommodation of local minima in watershed model calibration

The Gauss-Marquardt-Levenberg (GML) method of computer-based parameter estimation, in common with other gradient-based approaches, suffers from the drawback that it may become trapped in local objective function minima, and thus report optimized parameter values that are not, in fact, optimized at all. This can seriously degrade its utility in the calibration of watershed models where local optima abound. Nevertheless, the method also has advantages, chief among these being its model-run efficiency, and its ability to report useful information on parameter sensitivities and covariances as a by-product of its use. It is also easily adapted to maintain this efficiency in the face of potential numerical problems (that adversely affect all parameter estimation methodologies) caused by parameter insensitivity and/or parameter correlation. The present paper presents two algorithmic enhancements to the GML method that retain its strengths, but which overcome its weaknesses in the face of local optima. Using the first of these methods an intelligent search for better parameter sets is conducted in parameter subspaces of decreasing dimensionality when progress of the parameter estimation process is slowed either by numerical instability incurred through problem ill-posedness, or when a local objective function minimum is encountered. The second methodology minimizes the chance of successive GML parameter estimation runs finding the same objective function minimum by starting successive runs at points that are maximally removed from previous parameter trajectories. As well as enhancing the ability of a GML-based method to find the global objective function minimum, the latter technique can also be used to find the locations of many non-global optima (should they exist) in parameter space. This can provide a useful means of inquiring into the well-posedness of a parameter estimation problem, and for detecting the presence of bimodal parameter and predictive probability distributions. The new methodologies are demonstrated by calibrating a Hydrological Simulation Program-FORTRAN (HSPF) model against a time series of daily flows. Comparison with the SCE-UA method in this calibration context demonstrates a high level of comparative model run efficiency for the new method. (c) 2006 Elsevier B.V. All rights reserved.

The 'moving targets' training algorithm

A simple method for training the dynamical behavior of a neural network is derived. It is applicable to any training problem in discrete-time networks with arbitrary feedback. The algorithm resembles back-propagation in that an error function is minimized using a gradient-based method, but the optimization is carried out in the hidden part of state space either instead of, or in addition to weight space. Computational results are presented for some simple dynamical training problems, one of which requires response to a signal 100 time steps in the past.