Search

This paper introduces an IIR flexible ICA approach to the problem of blind source separation in convolutive nonlinear environments. The proposed algorithm performs separation in the presence of convolutive mixing of post nonlinear convolutive mixtures (CPNL-C), it is based on redundancy reduction and realizes source separation by minimizing the output mutual information. Experimental results are described to show the effectiveness of the described technique.

In this paper a new neural network model for blind demixing of nonlinear mixtures is proposed. We address the use of the adaptive spline neural network recently introduced for supervised and unsupervised neural networks. These networks are built using neurons with flexible B-spline activation functions and in order to separate signals from mixtures, a gradient-ascending algorithm which maximizes the outputs entropy is derived. In particular a suitable architecture composed by two layers of flexible nonlinear functions for the separation of nonlinear mixtures is proposed. Some experimental results that demonstrate the effectiveness of the proposed neural architecture are presented.

In this paper, making use of the signal-flow-graph (SFG) representation and its known properties, we derive a new general method for backward gradient computation of a system output or cost function with respect to past (or present) system parameters. The system can be any causal, in general nonlinear and time-variant dynamic system represented by a SFG, in particular any feedforward or recurrent neural network. In this work we use discrete time notation, but the same theory holds for the continuous time case. The gradient is obtained by the analysis of two SFGs, the original one and its adjoint. This method can be used both for online and off-line learning. In the latter case using the mean square error cost function, our approach particularises to Wan's method (1996) that is not suited for online training of recurrent networks. Computer simulations of nonlinear dynamic systems identification will also be presented to assess the performance of the algorithm resulting from the application of the proposed method in the case of locally recurrent neural networks.

Audio signal recovery is a common problem in the digital audio restoration field, because of corrupted samples that must be replaced. In this paper a subband architecture is presented for audio signal recovery, using neural nonlinear prediction based on adaptive spline neural networks. The experimental results show the mean square reconstruction error, and maximum error obtained with increasing gap length, from 200 to 5000 samples. The method gives good results allowing the reconstruction of over 100 ms of signal with low audible effects in overall quality.

A coarse-grained processor array which tries to overcome the limitations encountered when a network composed by many neurons has to be mapped into a limited number of processing elements (PEs) is proposed. Given the total number of PEs, the proposed architecture can be configured to implement any particular multilayer perceptron (MLP) topology, and can simulate both the forward and learning phases of the network. Moreover, it has a small number of very local connections, and can exhibit a high efficiency under limited constraints on the number of neurons per layer. >

The feasibility of restricting the weight values to powers-of-two or sums of powers-of-two in multilayer neural networks is discussed. A learning procedure based on back-propagation to obtain a neural network with such weights is presented. This learning procedure requires full real arithmetic, and therefore must be performed offline. These neural networks do not require multipliers, and are well suited for high-speed and high-integration digital neural circuits. To show the effectiveness of the approach, tests on a pattern recognition problem are presented. >

In this paper, we study the properties of neural networks based on adaptive spline activation functions. Using the results of regularization theory, we show how the proposed architecture is able to produce smooth approximations of unknown functions; to reduce hardware complexity a particular implementation of the kernels expected by the theory is suggested. This solution, although sub-optimal, greatly reduces the number of neurons and connections as it gives an increased expressive power to each neuron which is also able to produce a smooth activation function just controlling one fixed parameter of a Catmull-Rom cubic spline. Experimental results demonstrate that there is also an advantage in terms of the number of free parameters that, together with smoothness, leads to an improved generalization capability.

Over the last years, automatic music classification has become a standard benchmark problem in the machine learning community. This is partly due to its inherent difficulty, and also to the impact that a fully automated classification system can have in a commercial application. In this paper we test the efficiency of a relatively new learning tool, Extreme Learning Machines (ELM), for several classification tasks on publicly available song datasets. ELM is gaining increasing attention, due to its versatility and speed in adapting its internal parameters. Since both of these attributes are fundamental in music classification, ELM provides a good alternative to standard learning models. Our results support this claim, showing a sustained gain of ELM over a feedforward neural network architecture. In particular, ELM provides a great decrease in computational training time, and has always higher or comparable results in terms of efficiency.

In this paper a novel connectionist approach for video compression is presented. The basic idea is to extend to the temporal dimension the architecture used for the compression of still images. A multilayer perceptron (MLP) with infinite impulse response (IIR) synapses, embedded in a new quad-tree framework for video segmentation, is employed to take into account the video temporal dynamics. In order to reduce the computational burden and to improve the generalization performance, a flexible spline-based activation function, suitable for signal processing applications, has been used. Preliminary experimental results show that the proposed approach represents a viable alternative with respect to existing standards for high-quality video compression.

In this paper an alternative algorithm for the Singular Value Decomposition (SVD) of the data matrix used for Direction-Of-Arrival (DOA) estimation is presented. The proposed algorithm transforms the data matrix into a bi-diagonal form by a procedure based on fast Givens rotations. This procedure is specifically tailored for DOA estimation and allows a certain computational saving with respect to the well known Golub-Kahan algorithm. >

After an ad hoc extension of the multilayer perceptron model to complex signals, a blind neural network equalizer is proposed. Several experimental results are also reported which prove the effectiveness of this approach for equalizing a digital satellite radio channel in the presence of mild nonlinearities and intersymbol interference. Moreover, the proposed method works without any knowledge about the nonlinear nature of the channel. >

Presents a new neural architecture that is suitable for digital signal processing applications. The architecture, which is based on adaptable multidimensional activation functions, allows one to collect information from the previous network layer in aggregate form. In other words, the number of network connections (the structural complexity) can be very low with respect to the problem complexity. This fact, as experimentally demonstrated in this paper, improves the network's generalization capabilities and speeds up the convergence of the learning process. A specific learning algorithm is derived, and experimental results on channel equalization demonstrate the effectiveness of the proposed architecture.

It is well known that the behaviour of a neural network built with classical summing neurons, as in a multilayer perceptron, widely depends on the activation functions of the involved neurons. Many authors have proposed the use of activation functions with some free parameters which should allow one to reduce the size of the network, trading connection complexity with activation function complexity. Since many implementations of neural network are based on digital hardware, performing the selected activation function through a lookup-table (LUT), it could be interesting to study neural networks whose neurons have adaptable LUT-based activation functions. In this way, after learning, the neurons will present arbitrary activation functions which can also be efficiently implemented with digital technologies. In this paper a preliminary study of the adaptive LUT-based neuron (L-neuron) is presented, together with some experimental results on canonical problems.

This paper introduces a new framework for multichannel adaptive filtering, aiming at improving performance of an overall filtering system. The proposed architecture relies on the properties of the adaptive combination of filters which exploits the capabilities of different constituents, thus adaptively providing at least the behaviour of the best performing filter. Applying this concept to multichannel filtering systems, we define a scheme for the combination of multiple-input multiple-output (MIMO) filters. More precisely the proposed structure involves the combination of two different multiple-input singleoutput (MISO) systems for each MIMO output. We propose such framework with application to the multichannel acoustic echo cancellation (MAEC) with the goal of giving robustness to the system against impulsive background noise, and thus improving overall cancelling performance. Experimental results show the effectiveness of the proposed combined MAEC in the presence of adverse environmental conditions.