Self-organising neural networks for adaptive control

Self-organizing neural networks have been implemented in a wide range of application areas such as speech processing, image processing, optimization and robotics. Recent variations to the basic model proposed by the authors enable it to order state space using a subset of the input vector and to apply a local adaptation procedure that does not rely on a predefined test duration limit. Both these variations have been incorporated into a new feature map architecture that forms an integral part of an Hybrid Learning System (HLS) based on a genetic-based classifier system. Problems are represented within HLS as objects characterized by environmental features. Objects controlled by the system have preset targets set against a subset of their features. The system's objective is to achieve these targets by evolving a behavioural repertoire that efficiently explores and exploits the problem environment. Feature maps encode two types of knowledge within HLS — long-term memory traces of useful regularities within the environment and the classifier performance data calibrated against an object's feature states and targets. Self-organization of these networks constitutes non-genetic-based (experience-driven) learning within HLS. This paper presents a description of the HLS architecture and an analysis of the modified feature map implementing associative memory. Initial results are presented that demonstrate the behaviour of the system on a simple control task.

Using self-organising feature maps for the control of artificial organisms

Variations on the standard Kohonen feature map can enable an ordering of the map state space by using only a limited subset of the complete input vector. Also it is possible to employ merely a local adaptation procedure to order the map, rather than having to rely on global variables and objectives. Such variations have been included as part of a hybrid learning system (HLS) which has arisen out of a genetic-based classifier system. In the paper a description of the modified feature map is given, which constitutes the HLSs long term memory, and results in the control of a simple maze running task are presented, thereby demonstrating the value of goal related feedback within the overall network.

Neurofuzzy network model construction using Bézier-Bernstein polynomial functions

Neurofuzzy modelling systems combine fuzzy logic with quantitative artificial neural networks via a concept of fuzzification by using a fuzzy membership function usually based on B-splines and algebraic operators for inference, etc. The paper introduces a neurofuzzy model construction algorithm using Bezier-Bernstein polynomial functions as basis functions. The new network maintains most of the properties of the B-spline expansion based neurofuzzy system, such as the non-negativity of the basis functions, and unity of support but with the additional advantages of structural parsimony and Delaunay input space partitioning, avoiding the inherent computational problems of lattice networks. This new modelling network is based on the idea that an input vector can be mapped into barycentric co-ordinates with respect to a set of predetermined knots as vertices of a polygon (a set of tiled Delaunay triangles) over the input space. The network is expressed as the Bezier-Bernstein polynomial function of barycentric co-ordinates of the input vector. An inverse de Casteljau procedure using backpropagation is developed to obtain the input vector's barycentric co-ordinates that form the basis functions. Extension of the Bezier-Bernstein neurofuzzy algorithm to n-dimensional inputs is discussed followed by numerical examples to demonstrate the effectiveness of this new data based modelling approach.

Dual-orthogonal radial basis function networks for nonlinear time series prediction

A new structure of Radial Basis Function (RBF) neural network called the Dual-orthogonal RBF Network (DRBF) is introduced for nonlinear time series prediction. The hidden nodes of a conventional RBF network compare the Euclidean distance between the network input vector and the centres, and the node responses are radially symmetrical. But in time series prediction where the system input vectors are lagged system outputs, which are usually highly correlated, the Euclidean distance measure may not be appropriate. The DRBF network modifies the distance metric by introducing a classification function which is based on the estimation data set. Training the DRBF networks consists of two stages. Learning the classification related basis functions and the important input nodes, followed by selecting the regressors and learning the weights of the hidden nodes. In both cases, a forward Orthogonal Least Squares (OLS) selection procedure is applied, initially to select the important input nodes and then to select the important centres. Simulation results of single-step and multi-step ahead predictions over a test data set are included to demonstrate the effectiveness of the new approach.

Multi-level Input Fusion in multi-modal systems

Context-aware multimodal interactive systems aim to adapt to the needs and behavioural patterns of users and offer a way forward for enhancing the efficacy and quality of experience (QoE) in human-computer interaction. The various modalities that constribute to such systems each provide a specific uni-modal response that is integratively presented as a multi-modal interface capable of interpretation of multi-modal user input and appropriately responding to it through dynamically adapted multi-modal interactive flow management , This paper presents an initial background study in the context of the first phase of a PhD research programme in the area of optimisation of data fusion techniques to serve multimodal interactivite systems, their applications and requirements.

A study of the sensitivity of ocean overturning circulation and climate to freshwater input in different regions of the North Atlantic

The Atlantic meridional overturning circulation (AMOC) is an important component of the climate system. Models indicate that the AMOC can be perturbed by freshwater forcing in the North Atlantic. Using an ocean-atmosphere general circulation model, we investigate the dependence of such a perturbation of the AMOC, and the consequent climate change, on the region of freshwater forcing. A wide range of changes in AMOC strength is found after 100 years of freshwater forcing. The largest changes in AMOC strength occur when the regions of deepwater formation in the model are forced directly, although reductions in deepwater formation in one area may be compensated by enhanced formation elsewhere. North Atlantic average surface air temperatures correlate linearly with the AMOC decline, but warming may occur in localised regions, notably over Greenland and where deepwater formation is enhanced. This brings into question the representativeness of temperature changes inferred from Greenland ice-core records.

Understanding African easterly waves: a moist singular vector approach

Moist singular vectors (MSV) have been applied successfully to predicting mid-latitude storms growing in association with latent heat of condensation. Tropical cyclone sensitivity has also been assessed. Extending this approach to more general tropical weather systems here, MSVs are evaluated for understanding and predicting African easterly waves, given the importance of moist processes in their development. First results, without initial moisture perturbations, suggest MSVs may be used advantageously. Perturbations bear similar structural and energy profiles to previous idealised non-linear studies and observations. Strong sensitivities prevail in the metrics and trajectories chosen, and benefits of initial moisture perturbations should be appraised. Copyright © 2009 Royal Meteorological Society

What's in a name? Lexical knowledge drives infants' visual preferences in the absence of referential input

Seventeen-month-old infants were presented with pairs of images, in silence or with the non-directive auditory stimulus 'look!'. The images had been chosen so that one image depicted an item whose name was known to the infant, and the other image depicted an image whose name was not known to the infant. Infants looked longer at images for which they had names than at images for which they did not have names, despite the absence of any referential input. The experiment controlled for the familiarity of the objects depicted: in each trial, image pairs presented to infants had previously been judged by caregivers to be of roughly equal familiarity. From a theoretical perspective, the results indicate that objects with names are of intrinsic interest to the infant. The possible causal direction for this linkage is discussed and it is concluded that the results are consistent with Whorfian linguistic determinism, although other construals are possible. From a methodological perspective, the results have implications for the use of preferential looking as an index of early word comprehension.

A singular vector perspective of 4DVAR: The spatial structure and evolution of baroclinic weather systems

The extent to which the four-dimensional variational data assimilation (4DVAR) is able to use information about the time evolution of the atmosphere to infer the vertical spatial structure of baroclinic weather systems is investigated. The singular value decomposition (SVD) of the 4DVAR observability matrix is introduced as a novel technique to examine the spatial structure of analysis increments. Specific results are illustrated using 4DVAR analyses and SVD within an idealized 2D Eady model setting. Three different aspects are investigated. The first aspect considers correcting errors that result in normal-mode growth or decay. The results show that 4DVAR performs well at correcting growing errors but not decaying errors. Although it is possible for 4DVAR to correct decaying errors, the assimilation of observations can be detrimental to a forecast because 4DVAR is likely to add growing errors instead of correcting decaying errors. The second aspect shows that the singular values of the observability matrix are a useful tool to identify the optimal spatial and temporal locations for the observations. The results show that the ability to extract the time-evolution information can be maximized by placing the observations far apart in time. The third aspect considers correcting errors that result in nonmodal rapid growth. 4DVAR is able to use the model dynamics to infer some of the vertical structure. However, the specification of the case-dependent background error variances plays a crucial role.

A singular vector perspective of 4D-Var: Filtering and interpolation

Four-dimensional variational data assimilation (4D-Var) combines the information from a time sequence of observations with the model dynamics and a background state to produce an analysis. In this paper, a new mathematical insight into the behaviour of 4D-Var is gained from an extension of concepts that are used to assess the qualitative information content of observations in satellite retrievals. It is shown that the 4D-Var analysis increments can be written as a linear combination of the singular vectors of a matrix which is a function of both the observational and the forecast model systems. This formulation is used to consider the filtering and interpolating aspects of 4D-Var using idealized case-studies based on a simple model of baroclinic instability. The results of the 4D-Var case-studies exhibit the reconstruction of the state in unobserved regions as a consequence of the interpolation of observations through time. The results also exhibit the filtering of components with small spatial scales that correspond to noise, and the filtering of structures in unobserved regions. The singular vector perspective gives a very clear view of this filtering and interpolating by the 4D-Var algorithm and shows that the appropriate specification of the a priori statistics is vital to extract the largest possible amount of useful information from the observations. Copyright © 2005 Royal Meteorological Society

Study of the variation of the input impedance of induction machines with frequency

The results of a study of the variation of three-phase induction machines' input impedance with frequency are proposed. A range of motors were analysed, both two-pole and four-pole, and the magnitude and phase of the input impedance were obtained over a wide frequency range of 20 Hz-1 MHz. For test results that would be useful in the prediction of the performance of induction machines during typical use, a test procedure was developed to represent closely typical three-phase stator coil connections when the induction machine is driven by a three-phase inverter. In addition, tests were performed with the motor's cases both grounded and not grounded. The results of the study show that all induction machines of the type considered exhibit a multiresonant impedance profile, where the input impedance reaches at least one maximum as the input frequency is increased. Furthermore, the test results show that the grounding of the motor's case has a significant effect on the impedance profile. Methods to exploit the input impedance profile of an induction machine to optimise machine and inverter systems are also discussed.

Singular vector ensemble forecasting systems and the prediction of flow dependent uncertainty

The ECMWF ensemble weather forecasts are generated by perturbing the initial conditions of the forecast using a subset of the singular vectors of the linearised propagator. Previous results show that when creating probabilistic forecasts from this ensemble better forecasts are obtained if the mean of the spread and the variability of the spread are calibrated separately. We show results from a simple linear model that suggest that this may be a generic property for all singular vector based ensemble forecasting systems based on only a subset of the full set of singular vectors.

Input usage, output mix and industry deregulation: an analysis of the Australian dairy manufacturing industry

In this paper we estimate a Translog output distance function for a balanced panel of state level data for the Australian dairy processing sector. We estimate a fixed effects specification employing Bayesian methods, with and without the imposition of monotonicity and curvature restrictions. Our results indicate that Tasmania and Victoria are the most technically efficient states with New South Wales being the least efficient. The imposition of theoretical restrictions marginally affects the results especially with respect to estimates of technical change and industry deregulation. Importantly, our bias estimates show changes in both input use and output mix that result from deregulation. Specifically, we find that deregulation has positively biased the production of butter, cheese and powders.

Bayesian estimation and selection of nonlinear vector error correction models: The case of the sugar-ethanol-oil nexus in Brazil

Nonlinear adjustment toward long-run price equilibrium relationships in the sugar-ethanol-oil nexus in Brazil is examined. We develop generalized bivariate error correction models that allow for cointegration between sugar, ethanol, and oil prices, where dynamic adjustments are potentially nonlinear functions of the disequilibrium errors. A range of models are estimated using Bayesian Monte Carlo Markov Chain algorithms and compared using Bayesian model selection methods. The results suggest that the long-run drivers of Brazilian sugar prices are oil prices and that there are nonlinearities in the adjustment processes of sugar and ethanol prices to oil price but linear adjustment between ethanol and sugar prices.