The globalization of business where single products and services are designed, developed, and manufactured in many different countries signals a significant need for cost-effective and reliable information movement and management capabilities. Similarly, consumers are seeking technologies which will allow them to visit the Smithsonian, scan a book in the Oxford Library, and interview a Japanese monk for a school report, all from the comfort of home. A necessary ingredient for realizing this global society is a strong telecommunications infrastructure. Our paper describes some of the customer needs and technology advances that are causing a revolution in planning global telecommunications networks. We present a new telecommunications paradigm and study its impact in two key areas: multi-country network routing at both the traffic and facility levels, and global network robustness.

In this paper, we consider the polynomial time inferability from positive data for unions of two tree pattern languages. A tree pattern is a structured pattern known as a term in logic programming, and a tree pattern language is the set of all ground instances of a tree pattern. We present a polynomial time algorithm to find a minimal union of two tree pattern languages containing given examples. Our algorithm can be considered as a natural extension of Plotkin's least generalization algorithm, which finds a minimal single tree pattern language. By using this algorithm, we can realize a consistent and conservative polynomial time inference machine that identifies unions of two tree pattern languages from positive data in the limit.

As part of Hitachi's development of clean semiconductor processing equipment, the Fluids Modeling Group of the Mechanical Engineering Research Laboratory is developing a computer model, CONTAMINATE, for simulating contamination of wafers in chemical vapor deposition (CVD) systems. CONTAMINATE is based on a 2D implementation of the SIMPLER algorithm for simulating convection/diffusion transport processes. The new model includes modules for simulating fluid flow, heat transfer, chemical reactions, and gas-phase formation and deposition of clusters and particles. CONTAMINATE outputs property fields and estimates of various film quality indices. Using CONTAMINATE we simulated a SiH4: O2: N2 gas mixture at 300 K flowing over a wafer heated to 700 K. System pressures were varied from 1-100 torr and SiH4 pressures from 0.1 to 10 torr. Deposition characteristics are in qualitative agreement with actual systems and are summarized as follows: (1) No particles larger than 0.1µm deposited for any of the conditions tested. (2) Film damage occurred above 10 torr, but no damage occurred below 10 torr. (3) Increasing SiH4 pressure at constant system pressure eliminated particle deposition because particles grew large enought that thermophoresis blocked particle diffusion. (4) Conformal deposition of featured surfaces was achieved only at 1 torr. (5) Film thickness variation over the diameter of the wafer was 15% at 100 torr, 3% at 10 torr, and 1% at 1 torr.

An optical array imaging system is presented with basic experimental results. First, a remote object is illuminated with laser light at an angle and the reflected light is detected with an array sensor after interfering it with the reference light. This process is repeated by changing the illumination angle to collect a set of fringe patterns, which are A/D converted and stored in a harddisk in a computer. Then, the data are processed on a computer, first, to estimate the complex-amplitude object wave fields, second, to derive the eigenvector with the maximum eigenvalue for the correlation of the estimated object fields, and finally, to form an image of the object. The derivation of the eigenvector follows an iterative algorithm, which can be interpreted as the process of repeating backward wave propagation of the field between the two apertures illuminating and detecting laser light. The eigenvector field can be expected to backpropagate to focus at a point on the object with the maximum coefficient of reflection, so that a beam-steering operation is applied to the eigenvector to form an image of the object. The method uses only the information of the array data and the lateral spacings of the receiving array (CCD) elements. Hence, the method can give good images of objects even if the reference light is uncollimated with an unknown distorted wavefront, and even if the illuminating angles are imprecise in three dimensions. Basic experimental results clearly show the usefulness of the method.

We consider the role of equivalence queries in learning unknown concepts using membership and equivalence queries. Equivalence queries have the following two roles: (R1) indicating whether a learning algorithm has succeeded to learn the unknown concept and (R2) providing counterexamples. In this paper, we consider the learning using membership and equivalence queries but using only the (R2) part of equivalence queries. In order to gain an insight into the learning membership and equivalence queries but using only the (R2) part of equivalence queries, we define equivalence-detecting problem". Let C be a representation class which is polynomial time learnable using membership and equivalence queries. We show that if the equivalence-detecting problem for C is polynomial time solvable then C is polynomial time learnable using membership and equivalence queries without using (R1). Moreover, we show that under certain conditions, the two notions, polynomial time solvability of equivalence-detecting problem" and polynomial time learnability using membership and equivalence queries without using (R1)", are equivalent. For concrete examples, we prove that dfas are polynomial time learnable using membership and equivalence queries without using (R1) in the learning situation where the algorithm is informed the number of states of the minimum states dfa accepting the target set in advance. On the other hand, we show that the equivalence-detecting problem for dfas are not solvable in the learning situation where the algorithm can use no additional information. This result together with our main result shows that, in this learning situation, the (R1) part of equivalence queries are necessary to learn dfas using membership and equivalence queries.

A pattern is a finite string of constant symbols and variable symbols. The language of a pattern is the set of all strings obtained by substituting any nonnull constant string for each variable symbol in the pattern. The class of pattern languages was introduced by Angluin in 1979 as a concrete class which is inferable from positive data. In this paper, we consider the decision problem whether for given two patterns there is a containment relation between their languages, which was posed by Angluin and its decidability remains open. We give some sufficient conditions to make this problem decidable. We also introduce the notions of generalizations and minimal generalizations common to a set of patterns. We characterize the above open problem using the minimal generalization.

A learning algorithm for neural controllers based on random search is proposed. The method presents an attractive feature in comparison with the learning of neural controllers using the standard backpropagation method. Namely, in this approach the identification of the unknown plant becomes unnecessary because the parameters of the controller are determined by a trial and error process. This is a favorable feature particularly in cases in which the characteristics of the system are complicated and consequently the identification is difficult or impossible to perform at all. As application examples, the learning control of the pendulum system and the maze problem are shown.

This paper designs and evaluates highly parallel VLSI processors for real time 2-D state-space digital filters using hierarchical behavioral description language and synthesizer. The architecture of the 2-D state-space digital filtering system is a linear systolic array of homogeneous VLSI processors, each of which consists of eight processing elements (PEs) executing 1-D state-space digital filtering with multi-input and multi-output. Hierarchical behavioral description language and synthesizer are adopted to design and evaluate PE's and the VLSI processors. One 16 bit fixed-point PE executing a (4, 4)-th order 2-D state-space digital filtering is described on the basis of distributed arithmetic in about 1,200 steps by the description language and is composed of 15 K gates in terms of 2 input NAND gate. One VLSI processor which is a cascade connection of eight PEs is composed of 129 K gates and can be integrated into one 1515 [mm2] VLSI chip using 1 µm CMOS standard cell. The 2-D state-space digital filtering system composed of 128 VLSI processors at 25 MHz clock can execute a 1,0241,024 image in 1.47 [msec] and thus can be applied to real-time conventional video signal processing.

In this paper, we deal with inductive inference of an indexed family of recursive languages. We give two sufficient conditions for inductive inferability of an indexed family from positive data, each of which does not depend on the indexing of the family. We introduce two notions of finite cross property for a class of languages and a pair of finite tell-tales for a language. The former is a generalization of finite elasticity due to Wright and the latter consists of two finite sets of strings one of which is a finite tell-tale introduced by Angluin. The main theorem in this paper is that if any language of a class has a pair of finite tell-tales, then the class is inferable from positive data. Also, it is shown that any language of a class with finite cross property has a pair of finite tell-tales. Hence a class with finite cross property is inferable from positive data. Further-more, it is proved that a language has a finite tell-tale if and only if there does not exist any infinite cross sequence of languages contained in the language.

The hole-trapping and electron-trapping characteristics in dry oxides following various chemical cleanings have been studied using the avalanche injection method. The results indicated that hole trap density was almost the same for the chemical cleanings. Electron traps with two capture cross sections, σ, were observed. Electron traps with σ210-17 cm2 were found to be independent of the chemical cleaning, while those with σ410-19 cm2 to depend on the cleaning. Comparison with previous works indicated that electron traps with larger σ were related to Si-OH bonds. The other electron trap showed the increasing trapping rate with increasing the current density injected into oxide. This was explained by trap generation due to electron injection. A correlation between the density of generated electron traps and the amount of Al contamination on surfaces before dry oxidation was observed.

Syntax-directed editors have several advantages in editing programs because programming is guided by the syntax and free from syntax errors. Nevertheless, they are less popular than text editiors. One of the reason is that they force a priori specified editing structures on the user and do not allow him to use his own structure. ACE (Algorithmically Customizable syntax-directed Editor) provides a solution for this problem by using a technique of machine learning; ACE has a special function of customizing the grammar algorithmically and interactively based on the learning method for grammars from examples and queries. The grammar used in the editor is customized through interaction with the user so that the user can edit his program in a more familiar structure. The customizing function has been implemented based on the methods for learning of context-free grammars from structural examples, for which the correctness and the efficiency are proved formally. This guarantees the soundness and the efficiency of customization. Furthermore, ACE can be used as an algorithmic and interactive tool to design grammars, which is required for several purposes such as compiler design and pretty-printer design.

This paper deals with the problem of translating Japanese adnominal particles into English according to the idea of Example-Based Machine Translation (EBMT) proposed by Nagao. Japanese adnominal particles are important because: (1) they are frequent function words; (2) to translate them into English is difficult because their translations are diversified; (3) EBMT's effectiveness for adnominal particles suggests that EBMT is effective for other function words, e. g., prepositions of European languages. In EBMT, (1) a database which consists of examples (pairs of a source language expression and its target language translation) is prepared as knowledge for translation; (2) an example whose source expression is similar to the input phrase or sentence is retrieved from the example database; (3) by replacements of corresponding words in the target expression of the retrieved example, the translation is obtained. The similarity in EBMT is computed by the summation of the distance between words multiplied by the weight of each word. The authors' method differs from preceding research in two important points: (1) the authors utilize a general thesaurus to compute the distance between words; (2) the authors propose a weight which changes for every input. The feasibility of our approach has been proven through experiments concerning success rate.

In general, a two-dimensional display is defined by two orthogonal unit vectors. In developing the display, discriminant analysis has a shortcoming that the extracted axes are not orthogonal in general. First, in order to overcome the shortcoming, we propose discriminant analysis which provides an orthonormal system in the transformed space. The transformation preserves the discriminatory ability in terms of the Fisher criterion. Second, we present a necessary and sufficient condition that discriminant analysis in the original space provides an orthonormal system. Finally, we investigate the relationship between orthogonal discriminant analysis and the Karhunen-Loeve expansion in the original space.

The elementary formal system (EFS, for short) is a kind of logic program which directly manipulates character strings. This paper outlines in brief the authors' studies on algorithmic learning theory developed in the framework of EFS's. We define two important classes of EFS's and a new hierarchy of various language classes. Then we discuss EFS's as logic programs. We show that EFS's form a good framework for inductive inference of languages by presenting model inference system for EFS's in Shapiro's sense. Using the framework we also show that inductive inference from positive data and PAC-learning are both much more powerful than they have been believed. We illustrate an application of our theoretical results to Molecular Biology.

Optimal static load balancing problems in open BCMP queueing networks with state-independent arrival and service rates are studied. Their examples include optimal static load balancing in distributed computer systems and static routing in communication networks. We refer to the load balancing policy of minimizing the overall mean response (or sojourn) time of a job as the overall optimal policy. We show the conditions that the solutions of the overall optimal policy satisfy and show that the policy uniquely determines the utilization of each service center, the mean delay for each class and each path class, etc., although the solution, the utilization for each class, the mean delay for all classes at each service center, etc., may not be unique. Then we give tha linear relations that characterize the set whose elements are the optimal solutions, and discuss the condition wherein the overall optimal policy has a unique solution. In parametric analysis and numerical calculation of optimal values of performance variables we must ensure whether they can be uniquely determined.

Lexical-Functional Grammars (LFG's) were introduced to define the syntax of natural languages. In LFG's, each node of a derivation tree has some attributes. An LFG G consists of a context-free grammar (cfg) G0 called the underlying cfg of G and a description Pfs of constraints between the values of the attributes. Pfs can specify (1) constraints between the value of an attribute of a node and those of its children, and (2) constraints between the value of an attribute of a node called a controller and that of a node called its controllee. RLFG's were introduced as a subclass of LFG's. In RLFG's, only constraints between the value of an attribute of a node and those of its children can be specified. It is shown in this paper that the class of languages generated by RLFG's is equal to the class of recursively enumerable languages. Some restrictions on LFG's were proposed for the purpose of efficient parsing. Among them are (1) the condition called a valid derivation, and (2) the condition that the underlying cfg is cycle-free. For an RLFG G, if the production rules of the underlying cfg of G are of the form AaB or Aa for nonterminal symbols A, B and a terminal symbol a, then G is called an R-RLFG. Every R-RLFG satisfies the above restriction (1) and (2). It is also shown in this paper that the class of languages generated by R-RLFG's contains an NP-hard language, which means that parsing in deterministic polynomial time of LFG's is impossible in general (unless PNP) even if the above restrictions (1) and (2) are satisfied.

This paper extends an earlier study on the T-Model neural network to its collective computational properties. We present arguments that it is necessary to use the half-interconnected T-Model networks rather than the fully-interconnected Hopfield model networks. The T-Model has been generated in response to a number of observed weaknesses in the Hopfield model. This paper identities these problems and show how the T-Model overcomes them. The T-Model network is essentially a feedforward network which does not produce a local minimum for computations. A concept for understanding the dynamics of the T-Model neural circuit is presented and its performance is also compared with the Hopfield model. The T-Model neural circuit is implemented and tested with standard CMOS technology. Simulations and experiments show that the T-Model allows immense collective network computations and does not produce a local minimum. High densities comparable to that of the Hopfield model implementations have also been achieved.

In this paper, voltage-input current-output Membership Function Circuit (MFC) and Normalization Locked Loop (NLL) are proposed. They are useful building blocks for the current-mode analog fuzzy hardware. The voltage-input current-output MFC consists of one source coupled type Operational Transconductance Amplifier (OTA). The MFC is used in the input parts of the analog fuzzy hardware system. The fuzzy hardware system can execute the singleton fuzzy control algorithm. In the algorithm, the weighted average operation is processed. When the weighted average operation is directly realized by analog circuits, a divider must be implemented. Here, the NLL circuit, which can process the weighted average operation without the divider, is implemented using one source coupled type OTA. The proposed circuits were designed by using 2 µm CMOS design rules and its operations were confirmed using SPICE simulations.

This paper describes the design of a 38-GHz high power MMIC amplifier using an improved load-pull technique. We improved the load-pull technique accuracy by using MMIC transtormers to match the input and output impedances of a GaAs MESFET to about 50 ohms. We used this technique to measure the large-signal load impedance of a FET with a 600-µm-wide gate. Using the data obtained, we developed an MMIC amplifier composed of two of these FET cells. At 38 GHz, the amplifier has an output power of 23.5 dBm for a 1 dB gain compression level.

Fourier holographic image storage and reconstruction using BaTiO3 photorefractive crystal waveguide is investigated. The phase conjugation technique, which compensates image distortion caused by modal phase dispersion, successfully retores images stored in a test BaTiO3 crystal waveguide.