Showing papers in "Neurocomputing in 2002"

TL;DR: The methods of this paper are illustrated for RBF kernels and demonstrate how to obtain robust estimates with selection of an appropriate number of hidden units, in the case of outliers or non-Gaussian error distributions with heavy tails.

TL;DR: A modified version of support vector machines, called C-ascending support vector machine, is proposed to model non-stationary financial time series, where the recent e-insensitive errors are penalized more heavily than the distant e- insensitive errors.

TL;DR: This tutorial survey this subject with a principal focus on the most well-known models based on kernel substitution, namely, support vector machines.

TL;DR: In this paper, the authors studied the application of Hop"eld-like neural networks to optimization problems and proposed an efficient local minima avoidance strategy based on the continuous dynamics.

TL;DR: In this paper, the authors combine the wavelet decomposition (as a filtering step) and neural networks to provide an acceptable prediction value for nonlinear time series prediction, which results in a hierarchy of new time series that are easier to model and predict.

TL;DR: Off-line method exploits the linear–non-linear structure found in radial basis function neural networks to model the inside air temperature of a hydroponic greenhouse and its application to on-line learning is proposed.

TL;DR: It is shown that the use of the extended Kalman filter results in better learning than conventional RBF networks and faster learning than gradient descent.

TL;DR: A sequential learning algorithm is presented to adapt the structure of the network, in which it is possible to create a new hidden unit and also to detect and remove inactive units, using a pseudo-Gaussian function.

TL;DR: The main feature of this novel architecture is its capability of growing both in terms of map size as well as in a three-dimensional tree-structure in order to represent the hierarchical structure present in a data collection during an unsupervised training process, which makes it an ideal tool for data analysis and exploration.

TL;DR: It is shown how reactive agents can solve complex tasks without requiring any internal state and demonstrates that this is due to their ability to coordinate perception and action.

TL;DR: This work applied independent component analysis to concatenated single-trial EEG data from a fast go–nogo categorization task of natural images and showed that individual independent components might index neural synchrony within and between intracranial brain sources.

TL;DR: Experimental results show that a speed up ratio is achieved when applying a new approach to reduce the computation time taken by fast neural nets for the searching process to locate human faces automatically in cluttered scenes.

TL;DR: Four structures are developed for identifying different classes of nonlinear systems expressed in the input–output representation form by the adaptive time delay neural networks (ATDNN).

TL;DR: A procedure to generate di5erent scores for any candidate model from a single sample of training data is described and how to apply multiple comparison procedures (MCP) to model selection is discussed.

TL;DR: A new saliency map model with an adaptive masking method is proposed that generates the sequence of salient points from a color natural scene and a noise-tolerant generalized symmetry transformation is incorporated into the morphology operation.

TL;DR: Learning and Soft Computing (LearnSC) embodies 268 illustrations, 155 problems, 47 practical examples, 3 extended case studies on NNs based control, ÿnancial time series analysis, and computer graphics, as well as many sets of simulated experiments.

TL;DR: Simulations with synthetic evoked responses mixed into natural 122-channel MEG data show significant improvement in accuracy of signal restoration and the convex optimization problem is solved by a Newton-type method.

TL;DR: A physiologically constrained neural dynamical model of the visual system for the organization of attention and its mediation of object recognition and visual search, with the early visual areas playing a key role in mediating such interaction.

TL;DR: A data-driven deterministic clustering algorithm based on temporal cross-correlations and elements of graph theory to detect functionally connected regions is proposed and shows to successfully determine clusters related to the stimulus.

TL;DR: An improved method for achieving sparsity in least-squares support vector machines, which takes into account the residuals for all training patterns, rather than only those incorporated in the sparse kernel expansion.

TL;DR: It is seen that for this class of objective functions, the dimensionality of the problem is critical and with increasing numbers of decision variables, the learning becomes more and more difficult for ESs, and an “efficient” parameterization becomes crucial.

TL;DR: The architecture of radial basis function neural networks is modified so as to also model linear as well as the usual nonlinear input–output relationships, which is at least as powerful as the Takagi–Sugeno type of fuzzy rule-based systems.

TL;DR: A new image retrieval system for scenery images named IRIS (Image Retrieval by Impression words and Specific object names) which uses specific object names in addition to impression words which can reflect ambiguous human kansei (impression and sensitivity) as the retrieval keywords.

TL;DR: It is indicated that instantaneous mixing may hold in acoustic monitoring and a bilinear forms-based approach to instantaneous source separation is presented, showing that this method may give rise to a more robust separation.

TL;DR: An algorithm that provides adaptive plasticity in function approximation problems: the deformable (feature) map (DM) algorithm that is based on a self-organized deformation of the underlying multi-dimensional probability distributions is presented.

TL;DR: A dynamical classification method is introduced based on the principle of encoding sensory information in oscillating spatio-temporal patterns for the evaluation of EEG signals measured by spatially distributed electrodes over sensory cortices of rabbits and reveals two types of patterns in neocortex.

TL;DR: Simulations illustrate that the proposed method compares favorably to Hyvarinen's FastICA, Bell and Sejnowski's Infomax and Common's minimum of mutual information.

TL;DR: This paper demonstrates that, for the wide class of source distributions with certain non-null cumulants and a pre-specified scaling, separation is always a saddle point of a cumulant-based cost function and proposes a quasi-Newton approach for determining this saddle point.

TL;DR: Results show considerable improvement of the manufacturing quality obtained by the proposed nonlinear model predictive control of a simulated chaotic cutting process.

TL;DR: A system for the automatic segmentation of fluorescence micrographs is presented, in the first step, positions of fluorescent cells are detected by a fast learning neural network, which acquires the visual knowledge from a set of training cell-image patches selected by the user.