Classification of Random Binary Sequences Using Walsh-Fourier Analysis

The concept of a finite binary random sequence does not seem to be covered in the classical foundations of the theory of probability. Solomonoff, Kolmogorov and Chaitin have tried to include this case by considering the lengths of programs required to generate these sequences: a longer program implying more randomness. However this definition is difficult to apply. This paper presents a straightforward procedure using Walsh functions to determine the pattern in a binary sequence. A quantitative measure of randomness has also been proposed. This has been defined as the number of independent data (via the Walsh transform) required to generate the sequence divided by the length of the sequence. However at present this classification procedure is restricted to sequences of length 2k only. When extended to infinite sequences it yields results agreeing with those by the classical probability theory.     

New criterion of realisability

To check the physical realisability of a system, the Paley-Wiener criterion is usually employed. In the letter, an alternative criterion is put forward, which has the additional property of allowing the computation of the unrealisable part of the transfer function directly.     

A subsumptive, hierarchical, and distributed vision-based architecture for smart robotics

We present a distributed vision-based architecture for smart robotics that is composed of multiple control loops, each with a specialized level of competence. Our architecture is subsumptive and hierarchical, in the sense that each control loop can add to the competence level of the loops below, and in the sense that the loops can present a coarse-to-fine gradation with respect to vision sensing. At the coarsest level, the processing of sensory information enables a robot to become aware of the approximate location of an object in its field of view. On the other hand, at the finest end, the processing of stereo information enables a robot to determine more precisely the position and orientation of an object in the coordinate frame of the robot. The processing in each module of the control loops is completely independent and it can be performed at its own rate. A control Arbitrator ranks the results of each loop according to certain confidence indices, which are derived solely from the sensory information. This architecture has clear advantages regarding overall performance of the system, which is not affected by the "slowest link," and regarding fault tolerance, since faults in one module does not affect the other modules. At this time we are able to demonstrate the utility of the architecture for stereoscopic visual servoing. The architecture has also been applied to mobile robot navigation and can easily be extended to tasks such as "assembly-on-the-fly."     

Continuous unlearning in neural networks

An algorithm is described that improves the performance of Hopfield neural networks. This algorithm involves a continuous unlearning of spurious memories that goes together with the learning of the memories to be stored. The proposed algorithm is a variant of the Hebbian learning rule.<>     

Design Constraints and Reconstruction Algorithms for Traverse-Continuous-Rotate CT Scanners

In this paper a new configuration for a computerized tomographic (CT) scanner is presented. The machine is essentially a hybrid combination of second- and third-generation scanners. A single source of X-rays and an array of detectors are mounted on a gantry. The source/detector array grouping traverses the object while the gantry continuously rotates around the object. Conditions will be derived so that the projection data will completely cover the Radon space without any holes or partial overlaps, thus ensuring the existence of efficiently implementable reconstruction algorithms for inverting the data. We will also present a new convolution-backprojection algorithm for reconstructing tomographic images from data generated on such a scanner.     

Fast Classification Networks For Signal Processing

We present a generalization of the corner classification approach to training feedforward neural networks that allows rapid learning of nonbinary data. These generalized networks, called fast classification (FC) networks, are compared against backpropagation and radial basis function networks and are shown to have excellent performance for prediction of time series and pattern recognition. FC networks do not require iterative training and they can be used in many signal processing applications where fast, nonlinear filtering provides an advantage.     

Spectrum estimation with regularly missed observations

Jones and Parzen have studied the problem of spectral analysis of stationary normal time series with missing observations. The letter presents an alternative procedure which applies to nonnormal series as well. This procedure consists in extrapolating the observed samples into the missed-samples interval, and thereby estimating the new autocovariance and spectral-density functions.