Nabil H. Farhat

Optical Implementation Of The Hopfield Model

Optical implementation of content addressable associative memory based on the Hopfield model for neural networks and on the addition of nonlinear iterative feedback to a vector-matrix multiplier is described. Numerical and experimental results presented show that the approach is capable of introducing accuracy and robustness to optical processing while maintaining the traditional advantages of optics, namely, parallelism and massive interconnection capability. Moreover a potentially useful link between neural processing and optics that can be of interest in pattern recognition and machine vision is established.

Microwave diversity imaging and automated target identification based on models of neural networks

Radar targets can be identified by either forming images with sufficient resolution to be recognized by the human observer or by forming signatures or representations of the target for automated machine recognition. Tomographic microwave diversity imaging techniques that combine angular (aspect), spectral, and polarization degrees of freedom have been shown, as summarized in the first part of this paper, to be capable of producing images of the scattering centers of a target with near optical resolution. In the second part of the paper the author shows that collective nonlinear signal processing based on models of neural networks combined with the use of suitable target signatures (here sinogram representations) offer the promise of robust super-resolved target identification from partial information. Results presented are of numerical simulations for a neuromorphic processor where the neural net performs simultaneously the functions of data storage, processing, and recognition by automatically generating an identifying object label, and fast optoelectronic architectures and hardware implementations are briefly mentioned. Practical considerations and extensions to real systems are briefly discussed. >

Optoelectronic analogs of self-programming neural nets: architecture and methodologies for implementing fast stochastic learning by simulated annealing.

Self-organization and learning is a distinctive feature of neural nets and processors that sets them apart from conventional approaches to signal processing. It leads to self-programmability which alleviates the problem of programming complexity in artificial neural nets. In this paper architectures for partitioning an optoelectronic analog of a neural net into distinct layers with prescribed interconnectivity pattern to enable stochastic learning by simulated annealing in the context of a Boltzmann machine are presented. Stochastic learning is of interest because of its relevance to the role of noise in biological neural nets. Practical considerations and methodologies for appreciably accelerating stochastic learning in such a multilayered net are described. These include the use of parallel optical computing of the global energy of the net, the use of fast nonvolatile programmable spatial light modulators to realize fast plasticity, optical generation of random number arrays, and an adaptive noisy thresholding scheme that also makes stochastic learning more biologically plausible. The findings reported predict optoelectronic chips that can be used in the realization of optical learning machines.

Image understanding and interpretation in microwave diversity imaging

The authors investigate microwave imaging of metallic objects using a diversity method and interpret and predict the reconstructed image from an approach based on analysis of the scattering mechanism and a back-projection algorithm used in image retrieval. The connection between the various scattering mechanisms and the reconstructed images is discussed, what the images represent is interpreted, and a prediction is made as to what the image will look like over given spectral and angular windows. A brief description is given of the microwave diversity imaging system and the formulation of the microwave diversity imaging based on the physical optics approximation. The scattering mechanism of a complex shaped metallic object is then briefly reviewed and an alternate approach to interpreting the reconstruction image based on the understanding of the scattering mechanism and the reconstruction algorithm is given. Several numerical and experimental examples are included to support this interpretation approach. >

The bifurcating neuron network 2: an analog associative memory

The Bifurcating Neuron (BN), a chaotic integrate-and-fire neuron, is a model of a neuron augmented by coherent modulation from its environment. The BN is mathematically equivalent to the sine-circle map, and this equivalence relationship allowed us to apply the mathematics of one-dimensional maps to the design of a BN network. The study of the bifurcating diagram of the BN revealed that the BN, under a suitable condition, can function as an amplitude-to-phase converter. Also, being an integrate-and-fire neuron, it has an inherent capability to function as a coincidence detector. These two observations led us to the design of the BN Network 2 (BNN-2), a pulse-coupled neural network that exhibits associative memory of multiple analog patterns. In addition to the usual dynamical properties as an associative memory, the BNN-2 was shown to exhibit volume-holographic memory: it switches to different pages of its memory space as the frequency of the coherent modulation changes, meaning context-sensitive memory.

GBOPCAD: a synthesis tool for high-performance gain-boosted opamp design

A systematic design methodology for high-performance gain-boosted opamps (GBOs) is presented. The methodology allows the optimization of the GBO in terms of ac response and settling performance and is incorporated into an automatic computer-aided design (CAD) tool, called GBOPCAD. Analytic equations and heuristics are first used by GBOPCAD to obtain a sizing solution close to the global optimum. Then, simulated annealings are used by GBOPCAD to find the global optimum. A sample opamp is designed by this tool in a 0.6-/spl mu/m CMOS process. It achieves a dc gain of 80 dB, a unity-gain bandwidth of 836 MHz with 60/spl deg/ phase margin and a 0.0244% settling time of 5 ns. The sample/hold front-end of a 12-bit 50-MSample/s analog-digital converter was implemented with this opamp. It achieves a signal-to-noise ratio of 81.9 dB for a 8.1-MHz input signal.

Optimization of millimeter-wave glow-discharge detectors

A simplified theory that provides the basis for determining the optimum gas composition and pressure which maximize the responsivity of millimeter and submillimeter glow-discharge detectors at a given spectral frequency is presented together with experimental verification.

Analog realization of arbitrary one-dimensional maps

An increasing number of applications of a one-dimensional (1-D) map as an information processing element are found in the literature on artificial neural networks, image processing systems, and secure communication systems. In search of an efficient hardware implementation of a 1-D map, we discovered that the bifurcating neuron (BN), which was introduced elsewhere as a mathematical model of a biological neuron under the influence of an external sinusoidal signal, could provide a compact solution. The original work on the BN indicated that its firing time sequence, when it was subject to a sinusoidal driving signal, was related to the sine-circle map, suggesting that the BN can compute the sine-circle map. Despite its rich array of dynamical properties, the mathematical description of the BN is simple enough to lend itself to a compact circuit implementation. In this paper, we generalize the original work and show that the computational power of the BN can be extended to compute an arbitrary 1-D map. Also, we describe two possible circuit models of the BN: the programmable unijunction transistor oscillator neuron, which was introduced in the original work as a circuit model of the BN, and the integrated-circuit relaxation oscillator neuron (IRON), which was developed for more precise modeling of the BN. To demonstrate the computational power of the BN, we use the IRON to generate the bifurcation diagrams of the sine-circle map, the logistic map, as well as the tent map, and then compare them with exact numerical versions. The programming of the BN to compute an arbitrary map can be done simply by changing the waveform of the driving signal, which is given to the BN externally; this feature makes the circuit models of the BN especially useful in the circuit implementation of a network of 1-D maps.

Optoelectronic neural networks and learning machines

A brief neural-net primer based on phase-space and energy landscape considerations is presented. This provides the basis for subsequent discussion of optoelectronic architectures and implementations with self-organization and learning ability that are configured around an optical crossbar interconnect. Stochastic learning in the context of a Boltzmann machine is described to illustrate the flexibility of optoelectronics in performing tasks that may be difficult for electronics alone. Stochastic nets are studied to gain insight into the possible role of noise in biological neural nets. A description is given of two approaches to realizing large-scale optoelectronic neurocomputers: integrated optoelectronic neural chips with interchip optical interconnects that allow their clustering into large neural networks, and nets with a two-dimensional rather than one-dimensional arrangement of neurons and four-dimensional connectivity matrices for increased packing density and compatibility with two-dimensional data.<<ETX>>

Architectures for optoelectronic analogs of self-organizing neural networks.

Architectures for partitioning optoelectronic analogs of neural nets into input-output and internal groups to form a multilayered net capable of self-organization, self-programming, and learning are described. The architectures and implementation ideas given describe a class of optoelectronic neural net modules that, when interfaced to a conventional computer controller, can impart to it artificial intelligence attributes.