Prediction- and simulation-error based perceptron training: Solution space analysis and a novel combined training scheme

Previous papers have noted the difficulty in obtaining neural models which are stable under simulation when trained using prediction-error-based methods. Here the differences between series-parallel and parallel identification structures for training neural models are investigated. The effect of the error surface shape on training convergence and simulation performance is analysed using a standard algorithm operating in both training modes. A combined series-parallel/parallel training scheme is proposed, aiming to provide a more effective means of obtaining accurate neural simulation models. Simulation examples show the combined scheme is advantageous in circumstances where the solution space is known or suspected to be complex. (c) 2006 Elsevier B.V. All rights reserved.

Epipolar geometry estimation based on evolutionary agents

This paper presents a novel approach based on the use of evolutionary agents for epipolar geometry estimation. In contrast to conventional nonlinear optimization methods, the proposed technique employs each agent to denote a minimal subset to compute the fundamental matrix, and considers the data set of correspondences as a 1D cellular environment, in which the agents inhabit and evolve. The agents execute some evolutionary behavior, and evolve autonomously in a vast solution space to reach the optimal (or near optima) result. Then three different techniques are proposed in order to improve the searching ability and computational efficiency of the original agents. Subset template enables agents to collaborate more efficiently with each other, and inherit accurate information from the whole agent set. Competitive evolutionary agent (CEA) and finite multiple evolutionary agent (FMEA) apply a better evolutionary strategy or decision rule, and focus on different aspects of the evolutionary process. Experimental results with both synthetic data and real images show that the proposed agent-based approaches perform better than other typical methods in terms of accuracy and speed, and are more robust to noise and outliers.

A new Jacobian matrix for optimal learning of single-layer neural networks

This paper investigates the learning of a wide class of single-hidden-layer feedforward neural networks (SLFNs) with two sets of adjustable parameters, i.e., the nonlinear parameters in the hidden nodes and the linear output weights. The main objective is to both speed up the convergence of second-order learning algorithms such as Levenberg-Marquardt (LM), as well as to improve the network performance. This is achieved here by reducing the dimension of the solution space and by introducing a new Jacobian matrix. Unlike conventional supervised learning methods which optimize these two sets of parameters simultaneously, the linear output weights are first converted into dependent parameters, thereby removing the need for their explicit computation. Consequently, the neural network (NN) learning is performed over a solution space of reduced dimension. A new Jacobian matrix is then proposed for use with the popular second-order learning methods in order to achieve a more accurate approximation of the cost function. The efficacy of the proposed method is shown through an analysis of the computational complexity and by presenting simulation results from four different examples.

Strategies for Energy-Efficient Resource Management of Hybrid Programming Models

Many scientific applications are programmed using hybrid programming models that use both message passing and shared memory, due to the increasing prevalence of large-scale systems with multicore, multisocket nodes. Previous work has shown that energy efficiency can be improved using software-controlled execution schemes that consider both the programming model and the power-aware execution capabilities of the system. However, such approaches have focused on identifying optimal resource utilization for one programming model, either shared memory or message passing, in isolation. The potential solution space, thus the challenge, increases substantially when optimizing hybrid models since the possible resource configurations increase exponentially. Nonetheless, with the accelerating adoption of hybrid programming models, we increasingly need improved energy efficiency in hybrid parallel applications on large-scale systems. In this work, we present new software-controlled execution schemes that consider the effects of dynamic concurrency throttling (DCT) and dynamic voltage and frequency scaling (DVFS) in the context of hybrid programming models. Specifically, we present predictive models and novel algorithms based on statistical analysis that anticipate application power and time requirements under different concurrency and frequency configurations. We apply our models and methods to the NPB MZ benchmarks and selected applications from the ASC Sequoia codes. Overall, we achieve substantial energy savings (8.74 percent on average and up to 13.8 percent) with some performance gain (up to 7.5 percent) or negligible performance loss.

Optimal design of long-span steel portal frames using fabricated beams

This paper considers the optimal design of fabricated steel beams for long-span portal frames. The design optimisation takes into account ultimate as well as serviceability limit states, adopting deflection limits recommended by the Steel Construction Institute (SCI). Results for three benchmark frames demonstrate the efficiency of the optimisation methodology. A genetic algorithm (GA) was used to optimise the dimensions of the plates used for the columns, rafters and haunches. Discrete decision variables were adopted for the thickness of the steel plates and continuous variables for the breadth and depth of the plates. Strategies were developed to enhance the performance of the GA including solution space reduction and a hybrid initial population half of which is derived using Latin hypercube sampling. The results show that the proposed GA-based optimisation model generates optimal and near-optimal solutions consistently. A parametric study is then conducted on frames of different spans. A significant variation in weight between fabricated and conventional hot-rolled steel portal frames is shown; for a 50 m span frame, a 14–19% saving in weight was achieved. Furthermore, since Universal Beam sections in the UK come from a discrete section library, the results could also provide overall dimensions of other beams that could be more efficient for portal frames. Eurocode 3 was used for illustrative purposes; any alternative code of practice may be used.

An improved TLBO with elite strategy for parameters identification of PEM fuel cell and solar cell models

Clean and renewable energy generation and supply has drawn much attention worldwide in recent years, the proton exchange membrane (PEM) fuel cells and solar cells are among the most popular technologies. Accurately modeling the PEM fuel cells as well as solar cells is critical in their applications, and this involves the identification and optimization of model parameters. This is however challenging due to the highly nonlinear and complex nature of the models. In particular for PEM fuel cells, the model has to be optimized under different operation conditions, thus making the solution space extremely complex. In this paper, an improved and simplified teaching-learning based optimization algorithm (STLBO) is proposed to identify and optimize parameters for these two types of cell models. This is achieved by introducing an elite strategy to improve the quality of population and a local search is employed to further enhance the performance of the global best solution. To improve the diversity of the local search a chaotic map is also introduced. Compared with the basic TLBO, the structure of the proposed algorithm is much simplified and the searching ability is significantly enhanced. The performance of the proposed STLBO is firstly tested and verified on two low dimension decomposable problems and twelve large scale benchmark functions, then on the parameter identification of PEM fuel cell as well as solar cell models. Intensive experimental simulations show that the proposed STLBO exhibits excellent performance in terms of the accuracy and speed, in comparison with those reported in the literature.

Packet classification by multi-level cutting of the classification space: An algorithmic-architectural solution for IP packet classification in next generation networks

Traditionally, the Internet provides only a “best-effort” service, treating all packets going to the same destination equally. However, providing differentiated services for different users based on their quality requirements is increasingly becoming a demanding issue. For this, routers need to have the capability to distinguish and isolate traffic belonging to different flows. This ability to determine the flow each packet belongs to is called packet classification. Technology vendors are reluctant to support algorithmic solutions for classification due to their non-deterministic performance. Although CAMs are favoured by technology vendors due to their deterministic high lookup rates, they suffer from the problems of high power dissipation and high silicon cost. This paper provides a new algorithmic-architectural solution for packet classification that mixes CAMs with algorithms based on multi-level cutting the classification space into smaller spaces. The provided solution utilizes the geometrical distribution of rules in the classification space. It provides the deterministic performance of CAMs, support for dynamic updates, and added flexibility for system designers.

Packet classification by multi-level cutting of the classification space

Traditionally, the Internet provides only a “best-effort” service, treating all packets going to the same destination equally. However, providing differentiated services for different users based on their quality requirements is increasingly becoming a demanding issue. For this, routers need to have the capability to distinguish and isolate traffic belonging to different flows. This ability to determine the flow each packet belongs to is called packet classification. Technology vendors are reluctant to support algorithmic solutions for classification due to their nondeterministic performance. Although content addressable memories (CAMs) are favoured by technology vendors due to their deterministic high-lookup rates, they suffer from the problems of high-power consumption and high-silicon cost. This paper provides a new algorithmic-architectural solution for packet classification that mixes CAMs with algorithms based on multilevel cutting of the classification space into smaller spaces. The provided solution utilizes the geometrical distribution of rules in the classification space. It provides the deterministic performance of CAMs, support for dynamic updates, and added flexibility for system designers.

A hardware solution on multi-field packet classification

In this paper, a hardware solution for packet classification based on multi-fields is presented. The proposed scheme focuses on a new architecture based on the decomposition method. A hash circuit is used in order to reduce the memory space required for the Recursive Flow Classification (RFC) algorithm. The implementation results show that the proposed architecture achieves significant performance advantage that is comparable to that of some well-known algorithms. The solution is based on Altera Stratix III FPGA technology.

The List_Tree - an innovative solution to customer portfolio representation

The scale of BT's operations necessitates the use of very large scale computing systems, and the storage and management of large volumes of data. Customer product portfolios are an important form of data which can be difficult to store in a space efficient way. The difficulties arise from the inherently structured form of product portfolios, and the fact that they change over time as customers add or remove products. This paper introduces a new data-modelling abstraction called the List_Tree. It has been designed specifically to support the efficient storage and manipulation of customer product portfolios, but may also prove useful in other applications with similar general requirements.

Real-space grids and the Octopus code as tools for the development of new simulation approaches for electronic systems

Real-space grids are a powerful alternative for the simulation of electronic systems. One of the main advantages of the approach is the flexibility and simplicity of working directly in real space where the different fields are discretized on a grid, combined with competitive numerical performance and great potential for parallelization. These properties constitute a great advantage at the time of implementing and testing new physical models. Based on our experience with the Octopus code, in this article we discuss how the real-space approach has allowed for the recent development of new ideas for the simulation of electronic systems. Among these applications are approaches to calculate response properties, modeling of photoemission, optimal control of quantum systems, simulation of plasmonic systems, and the exact solution of the Schrödinger equation for low-dimensionality systems.