The Infona portal uses cookies, i.e. strings of text saved by a browser on the user's device. The portal can access those files and use them to remember the user's data, such as their chosen settings (screen view, interface language, etc.), or their login data. By using the Infona portal the user accepts automatic saving and using this information for portal operation purposes. More information on the subject can be found in the Privacy Policy and Terms of Service. By closing this window the user confirms that they have read the information on cookie usage, and they accept the privacy policy and the way cookies are used by the portal. You can change the cookie settings in your browser.
The development of neural network models requires the study of dedicated hardware architectures. In this paper, we propose on implementation of Radial Basis Function networks, derive an architecture based on an already existing 2D-systolic machine (MANTRA). A systolic algorithm is described to implement the required functions and the suitable sequence of operations. Theoretical efficiencies are estimated on the key tasks and some guidelines are given for a best usage of the Mantra machine in the studied framework...
This paper presents a texture segmentation model realised with image treatment processing and artificial neural network techniques. Gabor oriented filters are used to extract texture features and Self-Organising Feature Maps to group these features. In order to decrease the number of filters needed to better extract features, we use a multiresolution procedure and a learning rule property to features...
Learning neural algorithms are generally very simple, but the convergence is not very fast and robust. In this paper we address the important problem of optimum learning rate adjustement according to an adaptive procedure based on gradient method. The basic idea, very simple, which has already been successfully used in Signal Processing, is extended to 2 neural algorithms : Kohonen self-organizing...
In this paper we compare the implementations of Radial Basis Function (RBF) Neural Network on three parallel Neuro-Computers: the DRA machine (1D), the SMART machine (1D) and the MANTRA machine (2D). RBF networks can be used as probability density function estimators in a classification framework. The amount of calculation required for the simulation of such networks grows rapidly with the size of the learning database. Due to the highly parallel nature of RBF networks, parallel architectures are ideal candidates for such simulations. In this work we have tried to make a comparison of the three architectures based on the efficiency measure. We conclude this paper by outlining the different algorithmic constraints imposed by the particularities of each of the three architectures. We also discuss the I/O limitations for real time classification. Finally, we consider two real data-bases examples on which we compare the different machines...
As many countries, France recently set up an incentive policy for the installation of grid-connected PV-systems. In this context, it becomes important for an investor to be able to calculate the kWh-production from a given system in a specific site. The national program, Performance PV France, gathers research laboratories and French industrials in order to improve the PV system models and the solar...
This report proposesa2 microphone VoiceActivityDetector(VAD)and a Speech Enhancer (ENH) adapted to car conditions. The two modules are derived from the well-known Magnitude Square Coherence (MSC) which expresses a normalized cross-correlation for each frequency band of the received signals by the two sensors. A global VAD is directly obtained from the MSC by adaptive threshold which ensures a quasi-constant...
We present some recent progresses in the characterization of Ka-band radar return from water surfaces when observed at small incidence. We report on a radar experiment conducted at the large Pytheas wind-wave facility in Marseille and provide associated models. We consider the applicability of the results to natural oceanic and continental surfaces and show some specific properties of the sea clutter...
This document presents a two step analysis aiming at determining an optimised network of ground stations suitable for FSO satellite downlink communications. The optimisation is performed with respect to cloudiness for which experimental datasets are available. Daily visibility times or equivalently link capacities enabled by FSO are compared with X-band ones considering realistic GMES orbits. Both...
This paper presents an off-line method to estimate the mixing conditions, characterized by the number of audio sources and their time difference of arrival (TDOA). The proposed method is based on the assumption of having, at least, one time frame where each one of the sources is dominant. From such frames, the TDOA of the correspondent dominant source is estimated by maximizing a novel coherence function...
Hell in data analysis is paved (at least) with variability and noise. Is there some lost garden of Eden? Is there some way to approach it? In this paper, we deal with the human visual perception and we show how our visual system manages to process visual information in such a highly efficient way that it is able to categorize images or scenes within ranges of 100-150 ms independently of variability...
Set the date range to filter the displayed results. You can set a starting date, ending date or both. You can enter the dates manually or choose them from the calendar.