Developing maintainable expert systems using case-based reasoning

Abstract: Maintainability problems associated with traditional software systems are exacerbated in rule-based systems. The very nature of that approach — separation of control knowledge and data-driven execution — hampers maintenance. While there are widely accepted techniques for maintaining conventional software, the same is not true for rule-based systems. In most situations, both a knowledge engineer and a domain expert are necessary to update the rules of a rule-based system. This paper presents, first, an overview of the software engineering techniques and object-oriented methods used in maintaining rule-based systems. It then discusses alternate paradigms for expert system development. The benefits of using case-based reasoning (from the maintenance point of view) are illustrated through the implementation of a case-based scheduler. The main value of the scheduler is that its knowledge base can be modified by the expert without the assistance of a knowledge engineer. Since changes in application requirements can be given directly to the system by the expert, the effort of maintaining the knowledge base is greatly reduced.     

New type of charged-particles detector based on point contacts formed in a GaAs-AlGaAs heterostructure by split-gates

We propose a new principle for a low temperature semiconductor detector of charged particles with possibly submicron lateral resolution based on point contacts formed in the 2-dimensional electron gas (2DEG) of a GaAs-Al_xGa_1-xAs heterostructure using split-gates. The detector operates up to liquid nitrogen temperatures. Impinging particles excite locally extra donors in the doping layer of the heterostructure. By measuring simultaneous increase in the conductances of 3 point contacts due to the impacts of charged particles the position of penetration is calculated using the Thomas-Fermi approximation for screening in 2DEG.     

Hereditary neuralgic amyotrophy and hereditary neuropathy with liability to pressure palsies: two distinct clinical, electrophysiologic, and genetic entities

Fistulas secondary to Crohn's disease occur in about 33% of patients. A colobronchial fistula complicating Crohn's disease is very rare, having been reported only twice previously. We present an unusual fistula secondary to Crohn's colitis that originated from the splenic flexure and crossed the diaphragm to involve the bronchial tree.     

On Toroidal Azulenoids and Other Shapes of Fullerene Cage

The possibility of toroidal cages of carbon with repeating five and seven membered ring pairs (azulene units) upon their surfaces is briefly discussed. Some of the difficulties in recognising possible fullerene cage shapes are pointed out.     

Adaptive digital notch filter design on the unit circle for theremoval of powerline noise from biomedical signals

Investigates adaptive digital notch filters for the elimination of powerline noise from biomedical signals. Since the distribution of the frequency variation of the powerline noise may or may not be centered at 60 Hz. Three different adaptive digital notch filters are considered. For the first case, an adaptive FIR second-order digital notch filter is designed to track the center frequency variation. For the second case, the zeroes of an adaptive IIR second-order digital notch filter are fixed on the unit circle and the poles are adapted to find an optimum bandwidth to eliminate the noise to a pre-defined attenuation level. In the third case, both the poles and zeroes of the adaptive IIR second-order filter are adapted to track the center frequency variation within an optimum bandwidth. The adaptive process is considerably simplified by designing the notch filters by pole-zero placement on the unit circle using some suggested rules. A constrained least mean-squared algorithm is used for the adaptive process. To evaluate their performance, the three adaptive notch filters are applied to a powerline noise sample and to a noisy EEG as an illustration of a biomedical signal     

Design, manufacture and evaluation of a scalable high-performanceneural system

The authors describe a scalable neural system, HyperNet, based on a probabilistic RAM-based architecture and using a custom VLSI IC. A system using five HyperNet VLSI ICs and capable of realising up to 10240 neurons has been designed, manufactured and demonstrated to have the potential to learn more than three orders of magnitude faster than simulations on current workstations     

Addressing substrate coupling in mixed-mode ICs: simulation andpower distribution synthesis

This paper describes new techniques for the simulation and power distribution synthesis of mixed analog/digital integrated circuits considering the parasitic coupling of noise through the common substrate. By spatially discretizing a simplified form of Maxwell's equations, a three-dimensional linear mesh model of the substrate is developed. For simulation, a macromodel of the fine substrate mesh is formulated and a modified version of SPICE3 is used to simulate the electrical circuit coupled with the macromodel. For synthesis, a coarse substrate mesh, and interconnect models are used to couple linear macromodels of circuit functional blocks. Asymptotic Waveform Evaluation (AWE) is used to evaluate the electrical behavior of the network at every iteration in the synthesis process. Macromodel simulations are significantly faster than device level simulations and compare accurately to measured results. Synthesis results demonstrate the critical need to constrain substrate noise and simultaneously optimize power bus geometry and pad assignment to meet performance targets     

Excitation loss of open resonators

Reported here are the results on level and structure of the radiated field of an open resonator at and off resonance. On the basis of the modelling problem rigorous solution and the measurements at 4-millimeter wavelength we find there are no ways to compensate the radiated field from an open electromagnetic structure at resonance. Analysis of open resonator excitation efficiency shows that the mismatch loss of a slot-fed open resonator runs to 34% while at the aperture-feed excitation this is not more than 6% for the same-geometry resonator.