Millimeter-wave compressive holography.

We describe an active millimeter-wave holographic imaging system that uses compressive measurements for three-dimensional (3D) tomographic object estimation. Our system records a two-dimensional (2D) digitized Gabor hologram by translating a single pixel incoherent receiver. Two approaches for compressive measurement are undertaken: nonlinear inversion of a 2D Gabor hologram for 3D object estimation and nonlinear inversion of a randomly subsampled Gabor hologram for 3D object estimation. The object estimation algorithm minimizes a convex quadratic problem using total variation (TV) regularization for 3D object estimation. We compare object reconstructions using linear backpropagation and TV minimization, and we present simulated and experimental reconstructions from both compressive measurement strategies. In contrast with backpropagation, which estimates the 3D electromagnetic field, TV minimization estimates the 3D object that produces the field. Despite undersampling, range resolution is consistent with the extent of the 3D object band volume.

Second-Order Training of Adaptive Critics for Online Process Control

A new algorithm for training of nonlinear optimal neuro-controllers (in the form of the model-free, action-dependent, adaptive critic paradigm). Overcomes problems with existing stochastic backpropagation training: need for data storage, parameter shadowing and poor convergence, offering significant benefits for online applications.

Neural network controlled three-phase four-wire shunt active powerfilter

A three-phase four-wire shunt active power filter for harmonic mitigation and reactive power compensation in power systems supplying nonlinear loads is presented. Three adaptive linear neurons are used to tackle the desired three-phase filter current templates. Another feedforward three-layer neural network is adopted to control the output filter compensating currents online. This is accomplished by producing the appropriate switching patterns of the converter's legs IGBTs. Adequate tracking of the filter current references is obtained by this method. The active filter injects the current required to compensate for the harmonic and reactive components of the line currents, Simulation results of the proposed active filter indicate a remarkable improvement in the source current waveforms. This is reflected in the enhancement of the unified power quality index defined. Also, the filter has exhibited quite a high dynamic response for step variations in the load current, assuring its potential for real-time applications

A new Jacobian matrix for optimal learning of single-layer neural networks

This paper investigates the learning of a wide class of single-hidden-layer feedforward neural networks (SLFNs) with two sets of adjustable parameters, i.e., the nonlinear parameters in the hidden nodes and the linear output weights. The main objective is to both speed up the convergence of second-order learning algorithms such as Levenberg-Marquardt (LM), as well as to improve the network performance. This is achieved here by reducing the dimension of the solution space and by introducing a new Jacobian matrix. Unlike conventional supervised learning methods which optimize these two sets of parameters simultaneously, the linear output weights are first converted into dependent parameters, thereby removing the need for their explicit computation. Consequently, the neural network (NN) learning is performed over a solution space of reduced dimension. A new Jacobian matrix is then proposed for use with the popular second-order learning methods in order to achieve a more accurate approximation of the cost function. The efficacy of the proposed method is shown through an analysis of the computational complexity and by presenting simulation results from four different examples.

Improving neural network training solutions using regularisation

This paper describes the application of regularisation to the training of feedforward neural networks, as a means of improving the quality of solutions obtained. The basic principles of regularisation theory are outlined for both linear and nonlinear training and then extended to cover a new hybrid training algorithm for feedforward neural networks recently proposed by the authors. The concept of functional regularisation is also introduced and discussed in relation to MLP and RBF networks. The tendency for the hybrid training algorithm and many linear optimisation strategies to generate large magnitude weight solutions when applied to ill-conditioned neural paradigms is illustrated graphically and reasoned analytically. While such weight solutions do not generally result in poor fits, it is argued that they could be subject to numerical instability and are therefore undesirable. Using an illustrative example it is shown that, as well as being beneficial from a generalisation perspective, regularisation also provides a means for controlling the magnitude of solutions. (C) 2001 Elsevier Science B.V. All rights reserved.

The efficacy of colour cues in facilitating adaptation to opposing visuomotor rotations

We investigated visuomotor adaptation using an isometric, target-acquisition task. Following trials with no rotation, two participant groups were exposed to a random sequence of 30 degrees clockwise (CW) and 60 degrees counter-clockwise (CCW) rotations, with (DUAL-CUE), or without (DUAL-NO CUE), colour cues that enabled each environment (non-rotated, 30 degrees CW and 60 degrees CCW) to be identified. A further three groups experienced only 30 degrees CW trials or only 60 degrees CCW trials (SINGLE rotation groups) in which each visuomotor mapping was again associated with a colour cue. During training, all SINGLE groups reduced angular deviations of the cursor path during the initial portion of the movements, indicating feedforward adaptation. Consistent with the view that the adaptation occurred automatically via recalibration of the visuomotor mapping (Krakauer et al. 1999), post-training aftereffects were observed, despite colour cues that indicated that no rotation was present. For the DUAL-CUE group, angular deviations decreased with training in the 60 degrees trials, but were unchanged in the 30 degrees trials, while for the DUAL-NO CUE group angular deviations decreased for the 60 degrees CW trials but increased for the 30 degrees CW trials. These results suggest that in a dual adaptation paradigm a colour cue can permit delineation of the two environments, with a subsequent change in behaviour resulting in improved performance in at least one of these environments. Increased reaction times within the training block, together with the absence of aftereffects in the post-training period for the DUAL-CUE group suggest an explicit cue-dependent strategy was used in an attempt to compensate for the rotations.

The interference effects of non-rotated versus counter-rotated trials in visuomotor adaptation

An isometric torque-production task was used to investigate interference and retention in adaptation to multiple visuomotor environments. Subjects produced isometric flexion-extension and pronation-supination elbow torques to move a cursor to acquire targets as quickly as possible. Adaptation to a 30 degrees counter-clockwise (CCW) rotation (task A), was followed by a period of rest (control), trials with no rotation (task B0), or trials with a 60 degrees clockwise (CW) rotation (task B60). For all groups, retention of task A was assessed 5 h later. With initial training, all groups reduced the angular deviation of cursor paths early in the movements, indicating feedforward adaptation. For the control group, performance at commencement of the retest was significantly better than that at the beginning of the initial learning. For the B0 group, performance in the retest of task A was not dissimilar to that at the start of the initial learning, while for the B60 group retest performance in task A was markedly worse than initially observed. Our results indicate that close juxtaposition of two visuomotor environments precludes improved retest performance in the initial environment. Data for the B60 group, specifically larger angular errors upon retest compared with initial exposures, are consistent with the presence of anterograde interference. Furthermore, full interference occurred even when the visuomotor environment encountered in the second task was not rotated (B0). This latter novel result differs from those obtained for force field learning, where interference does not occur when task B does not impose perturbing forces, i.e., when B consists of a null field (Brashers-Krug et al., Nature 382:252-255, 1996). The results are consistent with recent proposals suggesting different interference mechanisms for visuomotor (kinematic) compared to force field (dynamic) adaptations, and have implications for the use of washout trials when studying interference between multiple visuomotor environments.

Optimum decision delay of the finite-length DFE

This letter investigates the optimum decision delay and tap-length of the finite-length decision feedback equalizer. First we show that, if the feedback filter (FBF) length N-b is equal to or larger than the channel memory upsilon and the decision delay Delta is smaller than the feedforward filter (FFF) length N-f, then only the first Delta + 1 elements of the FFF can be nonzero. Based on this result we prove that the maximum effective FBF length is equal to the channel memory upsilon, and if N-b greater than or equal to upsilon and N-f is long enough, the optimum decision delay that minimizes the MMSE is N-f - 1.

A self-structured adaptive decision feedback equalizer

In a decision feedback equalizer (DFE), the structural parameters, including the decision delay, the feedforward filter (FFF), and feedback filter (FBF) lengths, must be carefully chosen, as they greatly influence the performance. Although the FBF length can be set as the channel memory, there is no closed-form expression for the FFF length and decision delay. In this letter, first we analytically show that the two-dimensional search for the optimum FFF length and decision delay can be simplified to a one-dimensional search and then describe a new adaptive DFE where the optimum structural parameters can he self-adapted.

Split slot ring spatial quasi-circulator for RCS characterisation

A prototype X-band scale model for a quasi-optical three-port circulator utilising a double-layer circularly polarising frequency selective surface is proposed. The operating principles and measured characteristics of the device are discussed. A prototype device operating at 9.9 GHz has been built and validated experimentally. The port 1 to port 2 insertion loss of the quasi-circulator has been measured to be 2 dB, while port 1 to port 3 isolation is 16 dB. It is demonstrated that port 1 to 3 isolation can be increased to 25 dB by embedding the quasi-circulator in a feedforward setup.

Real-time error detection but not error correction drives automatic visuomotor adaptation

We investigated the role of visual feedback of task performance in visuomotor adaptation. Participants produced novel two degrees of freedom movements (elbow flexion-extension, forearm pronation-supination) to move a cursor towards visual targets. Following trials with no rotation, participants were exposed to a 60A degrees visuomotor rotation, before returning to the non-rotated condition. A colour cue on each trial permitted identification of the rotated/non-rotated contexts. Participants could not see their arm but received continuous and concurrent visual feedback (CF) of a cursor representing limb position or post-trial visual feedback (PF) representing the movement trajectory. Separate groups of participants who received CF were instructed that online modifications of their movements either were, or were not, permissible as a means of improving performance. Feedforward-mediated performance improvements occurred for both CF and PF groups in the rotated environment. Furthermore, for CF participants this adaptation occurred regardless of whether feedback modifications of motor commands were permissible. Upon re-exposure to the non-rotated environment participants in the CF, but not PF, groups exhibited post-training aftereffects, manifested as greater angular deviations from a straight initial trajectory, with respect to the pre-rotation trials. Accordingly, the nature of the performance improvements that occurred was dependent upon the timing of the visual feedback of task performance. Continuous visual feedback of task performance during task execution appears critical in realising automatic visuomotor adaptation through a recalibration of the visuomotor mapping that transforms visual inputs into appropriate motor commands.

Neural network forecasts of input-output technology

A significant part of the literature on input-output (IO) analysis is dedicated to the development and application of methodologies forecasting and updating technology coefficients and multipliers. Prominent among such techniques is the RAS method, while more information demanding econometric methods, as well as other less promising ones, have been proposed. However, there has been little interest expressed in the use of more modern and often more innovative methods, such as neural networks in IO analysis in general. This study constructs, proposes and applies a Backpropagation Neural Network (BPN) with the purpose of forecasting IO technology coefficients and subsequently multipliers. The RAS method is also applied on the same set of UK IO tables, and the discussion of results of both methods is accompanied by a comparative analysis. The results show that the BPN offers a valid alternative way of IO technology forecasting and many forecasts were more accurate using this method. Overall, however, the RAS method outperformed the BPN but the difference is rather small to be systematic and there are further ways to improve the performance of the BPN.

Spatial updating across saccades during manual interception

We studied the effect of intervening saccades on the manual interception of a moving target. Previous studies suggest that stationary reach goals are coded and updated across saccades in gaze-centered coordinates, but whether this generalizes to interception is unknown. Subjects (n = 9) reached to manually intercept a moving target after it was rendered invisible. Subjects either fixated throughout the trial or made a saccade before reaching (both fixation points were in the range of -10° to 10°). Consistent with previous findings and our control experiment with stationary targets, the interception errors depended on the direction of the remembered moving goal relative to the new eye position, as if the target is coded and updated across the saccade in gaze-centered coordinates. However, our results were also more variable in that the interception errors for more than half of our subjects also depended on the goal direction relative to the initial gaze direction. This suggests that the feedforward transformations for interception differ from those for stationary targets. Our analyses show that the interception errors reflect a combination of biases in the (gaze-centered) representation of target motion and in the transformation of goal information into body-centered coordinates for action.

Multi-scale keypoints in V1 and face detection

End-stopped cells in cortical area V1, which combine out- puts of complex cells tuned to different orientations, serve to detect line and edge crossings (junctions) and points with a large curvature. In this paper we study the importance of the multi-scale keypoint representa- tion, i.e. retinotopic keypoint maps which are tuned to different spatial frequencies (scale or Level-of-Detail). We show that this representation provides important information for Focus-of-Attention (FoA) and object detection. In particular, we show that hierarchically-structured saliency maps for FoA can be obtained, and that combinations over scales in conjunction with spatial symmetries can lead to face detection through grouping operators that deal with keypoints at the eyes, nose and mouth, especially when non-classical receptive field inhibition is employed. Al- though a face detector can be based on feedforward and feedback loops within area V1, such an operator must be embedded into dorsal and ventral data streams to and from higher areas for obtaining translation-, rotation- and scale-invariant face (object) detection.

Face segregation and recognition by cortical multi-scale line and edge coding

Models of visual perception are based on image representations in cortical area V1 and higher areas which contain many cell layers for feature extraction. Basic simple, complex and end-stopped cells provide input for line, edge and keypoint detection. In this paper we present an improved method for multi-scale line/edge detection based on simple and complex cells. We illustrate the line/edge representation for object reconstruction, and we present models for multi-scale face (object) segregation and recognition that can be embedded into feedforward dorsal and ventral data streams (the “what” and “where” subsystems) with feedback streams from higher areas for obtaining translation, rotation and scale invariance.