Guaranteed two-pass convergence for supervised and inferential learning.

We present a theoretical analysis of a version of the LAPART adaptive inferencing neural network. Our main result is a proof that the new architecture, called LAPART 2, converges in two passes through a fixed training set of inputs. We also prove that it does not suffer from template proliferation. For comparison, Georgiopoulos et al. have proved the upper bound n-1 on the number of passes required for convergence for the ARTMAP architecture, where n is the size of the binary pattern input space. If the ARTMAP result is regarded as an n-pass, or finite-pass, convergence result, ours is then a two-pass, or fixed-pass, convergence result. Our results have added significance in that they apply to set-valued mappings, as opposed to the usual supervised learning model of affixing labels to classes.

Acquiring rule sets as a product of learning in a logical neural architecture.

Envisioning neural networks as systems that learn rules calls forth the verification issues already being studied in knowledge-based systems engineering, and complicates these with neural-network concepts such as nonlinear dynamics and distributed memories. We show that the issues can be clarified and the learned rules visualized symbolically by formalizing the semantics of rule-learning in the mathematical language of two-valued predicate logic. We further show that this can, at least in some cases, be done with a fairly simple logical model. We illustrate this with a combination of two example neural-network architectures, LAPART, designed to learn rules as logical inferences from binary data patterns, and the stack interval network, which converts real-valued data into binary patterns that preserve the semantics of the ordering of real values. We discuss the significance of the formal model in facilitating the analysis of the underlying logic of rule-learning and numerical data representation. We provide examples to illustrate the formal model, with the combined stack interval/LAPART networks extracting rules from numerical data.

A deployed engineering design retrieval system using neural networks.

We describe a neural information retrieval system (NIRS), now in production within the Boeing Company, which has been developed for the identification and retrieval of engineering designs. Two-dimensional and three-dimensional representations of engineering designs are input to adaptive resonance theory (ART-1) neural networks to produce clusters of similar parts. The trained networks are then used to recall an appropriate cluster when queried with a new part design. This application is of great practical value to industry because it aids in the identification, retrieval, and reuse of engineering designs, potentially saving large amounts of nonrecurring costs. In this paper, we review the application, the neural architectures and algorithms, and then give the current status and the lessons learned in developing a neural network system for production use in industry.

A neural architecture for pattern sequence verification through inferencing.

LAPART, a neural network architecture for logical inferencing and supervised learning is discussed. Emphasizing its use in recognizing familiar sequences of patterns by verifying pattern pairs inferred from prior experience. It consists of interconnected adaptive resonance theory (ART) networks. The interconnects enable LAPART to learn to infer one pattern class from another to form a predictive sequence. It predicts a next pattern class based upon recognition of a current pattern and tests the prediction as new data become available. A confirmed prediction aids verification of a familiar sequence, and a disconfirmation flags a novel pairing of patterns. A simulation of LAPART is applied to verification of a hypothetical, known target using a sequence of sensor images obtained along a predetermined approach path. Application issues are addressed with a simple strategy, and it is shown how they could be addressed in a more complete fashion. Other topics, including a logical interpretation of ART and LAPART, are discussed.

An optoelectronic implementation of the adaptive resonance neural network.

A solution to the problem of implementation of the adaptive resonance theory (ART) of neural networks that uses an optical correlator which allows the large body of correlator research to be leveraged in the implementation of ART is presented. The implementation takes advantage of the fact that one ART-based architecture, known as ART1, can be broken into several parts, some of which are better to implement in parallel. The control structure of ART, often regarded as its most complex part, is actually not very time consuming and can be done in electronics. The bottom-up and top-down gated pathways, however, are very time consuming to simulate and are difficult to implement directly in electronics due to the high number of interconnections. In addition to the design, the authors present experiments with a laboratory prototype to illustrate its feasibility and to discuss implementation details that arise in practice. This device can potentially outperform alternative implementations of ART1 by as much as two to three orders of magnitude in problems requiring especially large input fields.

Photorefractive adaptive resonance neural network.

We describe a novel adaptive resonance theory (ART) device that is fully optical in the input-output processing path. This device is based on holographic information processing in a photorefractive crystal. This sets up an associative pattern retrieval in a resonating loop that uses angle-multiplexed reference beams for pattern classification. A reset mechanism is used to reject any given beam, permitting an ART search strategy. The design is similar to an existing nonlearning optical associative memory, but ours permits learning and makes use of information that the other device discards. It is a suitable response to the challenges of connectivity, learning, and reset presented by ART architectures. Furthermore, the design includes an efficient mechanism for area normalization of templates. It also permits the user to capitalize on the ability of ART networks to process large patterns. This new device is expected to offer higher information storage density than alternative ART implementations.

Limitations of a class of binary phase-only filters.

The feasibility of using certain types of binary phase-only filter (BPOF) is investigated. A critical aspect of correlation filters is not often addressed in research on BPOF's: how well do they perform as classifiers in the presence of imperfectly matching templates? It is not enough to detect a single given signal in the presence of noise; it is equally critical to make the correct classification among a number of possible templates with a low false-alarm rate. We show that (+1, -1)-valued BPOF's based on the real part of a conventional matched filter can cause misclassification of simple patterns, even in the absence of noise. These are known to be suboptimal, but the seriousness of their limitations illustrates an important design issue. It is therefore concluded that other types of filters must be used for correlator-based neural network implementations and image processing in general. We also include a commentary on the potential for facing this type of problem with general POF's and BPOF's. The theoretical results are supported by computer simulation and optical experiments.

Hybrid optoelectronic adaptive resonance theory neural processor, ART1.

For industrial use, adaptive resonance theory (ART) neural networks have the potential of becoming an important component in a variety of commercial and military systems. Efficient software emulations of these networks are adequate in many of today's low-end applications such as information retrieval or group technology. But for larger applications, special-purpose hardware is required to achieve the expected performance requirements. Direct electronic implementation of this network model has proven difficult to scale to large-input dimensionality owing to the high degree of interconnectivity between layers. Here, a new hardware implementation design of ART1 is proposed that handles input dimensions of practical size. It efficiently combines the advantages of optical and electronic devices to produce a stand-alone ART1 processor. Parallel computations are relegated to free-space optics, while serial operations are performed in VLSI electronics. One possible physical realization of this architecture is proposed. No hardware has as yet been built.

Analysis of ventricular shape by echocardiography in normal fetuses, newborns, and infants.

Qualitative and quantitative changes in left ventricular shapes were analyzed in 14 normal fetuses, 29 normal newborns, and 12 normal infants. Qualitative observations demonstrated that most fetuses and newborns with dominant right ventricles had flattened or even indented interventricular septae, which changed left ventricular shape into an ellipse. In contrast, left ventricular shapes in infants were round, similar to shapes described in older children and adults. When changes in shape or septal distortions were gross, interobserver agreement was 100%; when changes were less altered from a circular shape, interobserver agreement was 78%. To avoid subjective misinterpretations, quantitative analyses were performed, including M mode echocardiographic comparisons of right ventricular/left ventricular dimensions and left ventricular cavity anterior-posterior/lateral diameters, as well as Fourier analysis of digitized tracings of the entire left ventricular shape. The right ventricular/left ventricular ratio, determined by M mode echocardiography, showed significant differences between fetuses (1.07 +/- 0.07) and newborns (0.62 +/- 0.12) (p less than .001). Infants had a significantly lower right ventricular/left ventricular ratio (0.45 +/- 0.01) when compared with newborns (p less than .01). Ratios of left ventricular anterior-posterior/lateral diastolic diameters were significantly lower (p less than .001) in newborns (0.66 +/- 0.08) when compared with those of infants (0.82 +/- 0.10). All diameters tended to increase (toward roundness) with systole and with aging. Fourier analysis allowed evaluation of the entire left ventricle, including that portion of the septum that qualitatively appeared most indented and could not be analyzed by either of the above techniques. Shape factor derived from idealized shapes ranging from a circle to an indented ellipse allowed comparison with digitized left ventricular tracings. This technique allowed accurate quantitation of the observed changes in shape. Fetuses had the highest diastolic shape factor (7.47 +/- 0.92), whereas infants' shape factors were lowest (2.12 +/- 0.41). A tendency toward roundness and loss of distortion occurred with aging. Systolic shape factor was lower with aging in each group studied. The Fourier technique used in this study allows evaluation of an arbitrarily large number of components of a shape, and thus a complete description of that shape is permitted. Comparisons of right ventricular/left ventricular diastolic dimensions and left ventricular anterior-posterior/lateral comparisons are subsets of this technique, which allow evaluation of only two points (circular component-first harmonic) or four points (elliptical component-second harmonic) of an overall shape.(ABSTRACT TRUNCATED AT 400 WORDS)

Silicone rubber applied within the eye: a preliminary study.

Clear, room temperature vulcanizing silicone rubber (RTV) has already been shown to have well-defined optical and mechanical properties that allow its use as an active optical element. In this work RTV is considered as a possible material for substitution of damaged and/or diseased parts of the human eye, such as the cornea and crystalline lens. The interaction of cured and uncured RTV with eye tissue and eyelike tissue was investigated, and the results support the use of this material in the eye.

Surface profile measurement and angular deflection monitoring using a scanning laser beam: a noncontact method.

A noncontact method for measuring the surface profile of a reflecting surface is presented. The scheme is based on the measurement of the phase difference between a surface reflected laser beam and a reference beam. Both beams are scanned past a photodetector by means of a rotating beam splitter. The slope of the surface at various points is electronically obtained, and then the surface profile in a plane is determined by direct integration. The accuracy of the method falls between the more elaborate interferometric techniques and mechanical contact schemes. A complete theoretical analysis of the method is presented as well as the results of various experimental tests.

Shearing interferometer as an interferometric filter for the reduction of scattered light.

The background radiation problem encountered in astronomical observations is reviewed, and it is observed that a considerable reduction of the background radiation can be achieved by using interferometric techniques. A longitudinal radial shearing interferometer is described which meets the requirements of this application. The formal development of its properties as well as the results of preliminary tests are presented. It is shown that for a small region of the telescope field, this interferometer can be expected to reduce the background radiation originating in the telescope by a factor of 50 to 500, to reduce the background radiation produced by Mie type scattering external to the telescope to a level below that produced by Rayleigh scattering, and to reduce the background radiation from Rayleigh scattering by a factor of two. These properties can be extended to broadband spectral operation and can be achieved without degrading the astrometric qualities of the telescope. The size of the interferometer is quite small making it easy to add to an existing telescope.