Applications of neural networks are prevailing in speech recognition research. In this paper, first, suitable role of neural network (mainly back-propagation based multi-layer type) in speech recognition, is discussed. Considering that speech is a long, variable length, structured pattern, a direction, in which neural network is used in cooperation with existing structural analysis frameworks, is recommended. Activities are surveyed, including those intended to cooperatively merge neural networks into dynamic programming based structural analysis framework. It is observed that considerable efforts have been paid to suppress the high nonlinearity of network output. As far as surveyed, no experiment in real field has been reported.

Threre are two types of nonlinear associative silicon retinas. One is a sparse Hopfield type neural network which is called a H-type retina and the other is its dual network which is called a DH-type retina. The input information sequences of H-type and HD-type retinas are given by nodes and links as voltages and currents respectively. The error correcting capacity (minimum basin of attraction) of H-type and DH-type retinas is decided by the minimum numbers of links of cutset and loop respectively. The operation principle of the regeneration is based on the voltage or current distribution of the neural field. The most important nonlinear operation in the retinas is a dynamic quantization to decide the binary value of each neuron output from the neighbor value. Also, the edge is emphasized by a line-process. The rates of compression of H-type and DH-type retinas used in the simulation are 1/8 and (2/3) (1/8) respectively, where 2/3 and 1/8 mean rates of the structural and binarizational compression respectively. We could have interesting and significant simulation results enough to make a chip.

This paper introduces a modeling of the human rotation invariant recognition mechanism at the neural level. In the model, mechanisms of memory search and mental rotation are realized in the process of minimizing the energy of a bi-directional connection network. The thrust of the paper is to explain temporal mental activities such as successive memory retrievals and continuous mental rotation in terms of state transitions of collective neurons based on nonequilibrium dynamics. We conclude that regularities emerging in the dynamics of intermittent chaos lead the recognition process in a structural and meaningful way.

This paper reviews the capability of the three layer neural network (TLNN) with one output neuron. The input set is restricted to a finite subset S of En, and the TLNN implements a function F such as F : S I={1, -1}, i,e., F is a dichotomy of S. How many functions (dichotomies) can it compute by appropriately adjusting parameters in the TLNN? Brief historical review, some theorems on the subject obtained so far, and related topics are presented. Several open problems are also included.

Principal Component Analysis (PCA) is a useful technique in feature extraction and data compression. It can be formulated as a statistical constrained maximization problem, whose solution is given by unit eigenvectors of the data covariance matrix. In a practical application like image compression, the problem can be solved numerically by a corresponding gradient ascent maximization algorithm. Such on-line algoritms can be good alternatives due to their parallelism and adaptivity to input data. The algorithms can be implemented in a local and homogeneous way in learning neural networks. One example is the Subspace Network. It is a regular layer of parallel artificial neurons with a learning rule that is completely homogeneous with respect to the neurons. However, due to the complete homogeneity, the learning rule does not converge to the unique basis given by the dominant eigenvectors, but any basis of this eigenvector subspace is possible. In many applications like data compression, the subspace is not sufficient but the actual eigenvectors or PCA coefficient vectors are needed. A new criterion, called the Weighted Subspace Criterion, is proposed, which makes a small symmetry-breaking change to the Subspace Criterion. Only the true eigenvectors are solutions. Making the corresponding change to the learning rule of the Subspace Network gives a modified learning rule, which can be still implemented on a homogeneous network architecture. In learning, the weight vectors will tend to the true eigenvectors.

This paper describes the configuration and performance of a stable, high compression video coding scheme suitable for broadcast quality. This scheme was developed for application to high quality image packet transmission in Asynchronous Transfer Mode (ATM) networks. There are two problems in implementing image packet transmission in ATM networks, namely the achievement of a compression scheme with high coding efficiency, and the achievement of an effective compensation method for cell loss. We describe a scheme which resolves both these problems. It comprises the division of a two-dimensional spectral image signal into several sub-bands. In the case of the high frequency band, block-matching interframe prediction and Discrete Cosine Transform (DCT) are applied to achieve high compression ratio, while intraframe DCT coding is applied to the baseband. This scheme, moreover, provides a stable compensation for cell loss. It is shown that, based on this system, an original image signal of 216Mbit/s is compressed to about 1/10, and a high quality reconstructed image stable to cell loss is obtained.

The Batcher banyan network is well known as a non-blocking switching fabric. However, it is conflict free only when there is no packets for the same destination. To cope with the arbitrary combination of packets, an additional network or special control sequence which causes the increase of the hardware or performance degradation is required. A Batcher Double Omega network with Combining (BDOC) is an elegant solution of this problem. It consists of a Batcher sorter and two double sized Omega networks. Like in the Batcher banyan network, packets are sorted by the destination label in the Batcher sorter. In the first Omega network called the distributer, a packet is routed by a tag corresponding to the sum of the label at the output of the Batcher sorter and the destination label. In the second (Inverse) Omega network called the concentrator, the original destination label is used as the routing tag, and packets are routed without any conflict. The BDOC is useful for an interconnection network to connect processors and memory modules in multiprocessor. Unlike conventional multistage interconnection networks for multiprocessors, packets are transferred in a serial and synchronized manner. The simple structure of the switching element enables a high speed operation which reduces the latency caused by the serial communication. Using the pipelined circuit switching, the address and data packets share the same control signal, and the structure of the switching element is much simplified. Moreover, packets combining which avoids the hot spot contention is realized easily in the concentrator.

A theoretical conjecture on fractal dimensions of a dendrite distribution in neural networks is presented on the basis of the dendrite tree model. It is shown that the fractal dimensions obtained by the model are consistent with the recent experimental data.

Artificial neurons and neural networks have been shown to perform Principal Component Analysis (PCA) when gradient ascent learning rules are used, which are related to the constrained maximization of statistical objective functions. Due to their parallelism and adaptivity to input data, such algorithms and their implementations in neural networks are potentially useful in feature extraction and data compression. In the companion paper(9), two such learning rules were derived from two criteria, the Subspace Criterion and the Weighted Subspace Criterion. It was shown that the only solutions to the latter problem are dominant eigenvectors of the data covariance matrix, which are the basis vectors of PCA. It was suggested by a simulation that the corresponding learning algorithm converges to these eigenvectors. A homogeneous neural network implementation was proposed for the algorithm. The learning algorithm is analyzed here in detail and it is shown that it can be approximated by a continuous-time differential equation that is obtained by averaging. It is shown that the asymptotically stable limits of this differntial equation are the eigenvectors. The neural network learning algorithm is further extended to a case in which each neuron has a sigmoidal nonlinear feedback activity function. Then no parameters specific to each neuron are needed, and the learning rule is fully homogeneous.

Information geometry is a new powerful method of information sciences. Information geometry is applied to manifolds of neural networks of various architectures. Here is proposed a new theoretical approach to the manifold consisting of feedforward neural networks, the manifold of Boltzmann machines and the manifold of neural networks of recurrent connections. This opens a new direction of studies on a family of neural networks, not a study of behaviors of single neural networks.

This paper deals with the problem of assigning tasks to the processors of a multiprocessor system such that the sum of execution and communication costs is minimized. If the number of processors is two, this problem can be solved efficiently using the network flow approach pioneered by Stone. This problem is, however, known to be NP-complete in the general case, and thus intractable for systems with a large number of processors. In this paper, we propose a network flow approach for the task assignment problem in homogeneous hypercube networks, i.e., hypercube networks with functionally identical processors. The task assignment problem for an n-dimensional homogeneous hypercube network of N (=2n) processors and M tasks is first transformed into n two-terminal network flow problems, and then solved in time no worse than O(M3 log N) by applying the Goldberg-Tarjan's maximum flow algorithm on each two-terminal network flow problem.

New methods are needed for accessing very large information services. This paper proposes the use of a user model neural network to allow better access to a news service. The network is constructed on the basis of articles read, and articles marked as rejected. It adapts over time to better represent the user's interests and rank the articles supplied by the news service. Using an augmented keyword search we can also search for articles using keywords in conjunction with the user model neural network. Trials of the system in a USENET news environment show promising results for the use of this approach in information retrieval.

By close analogy of annealing for solids, we devise a new algorithm, called APS, for the time evolution of both the state and the synapses of the Hopfield's neural network. Through constrainedly random perturbation of the synapses of the network, the evolution of the state will ignore the tremendous number of small minima and reach a good minimum. The synapses resemble the microstructure of a network. This new algorithm anneals the microstructure of the network through a thermal controlled process. And the algorithm allows us to obtain a good minimum of the Hopfield's model efficiently. We show the potential of this approach for optimization problems by applying it to the will-known traveling salesman problem. The performance of this new algorithm has been supported by many computer simulations.

First, a number of issues pertaining to analog VLSI implementation of Backpropagation (BP) and Deterministic Boltzmann Machine (DBM) learning algorithms are clarified. According to the results from software simulation, a mismatch between the activation function and derivative generated by independent circuits degrades the BP learning performance. The perfomance can be improved, however, by adjusting the gain of the activation function used to obtain the derivative, irrespective of the original activation function. Calculation errors embedded in the circuits also degrade the learning preformance. BP learning is sensitive to offset errors in multiplication in the learning process, and DBM learning is sensitive to asymmetry between the weight increment and decrement processes. Next, an analog VLSI architecture for implementing the algorithms using common building block circuits is proposed. The evaluation results of test chips confirm that synaptic weights can be updated up to 1 MHz and that a resolution exceeding 14 bits can be attained. The test chips successfully perform XOR learning using each algorithm.

It is important to develop methods of measuring radiated electromagnetic interference level that will produce identical results at all measuring locations. We have considered a number of problems which prevent the achievement of identical results, and proposed some solutions. However, agreement of measurement values adequate for practical purposes has not been achieved. After our successive studies, we finally became aware that there is a causal relationship with changes in the line-to-ground impedance of the power supply. It is presumed that power cables of AC-powered devices operate as antenna elements that produce emission. Thus changes in the power line-to-ground impedance cause variations in the radiation efficiency to produce a different EMI level. We therefore made plans to measure the values of line-to-ground impedance at the AC power outlet for the frequency range of 100kHz to 500MHz at various locations where measurements are made of EMI from EUT (Equipment Under Test). The impedance varies greatly between 6ohms and 2 k-ohm, not only according to the frequency, but also according to the measurement location. In such cases, the EMI level shows a different value even with the same EUT, and it usually increases-especially for vertical polarization. We have developed a new type of LISN (Line Impedance Stabilization Network or Artificial Mains Network) to stabilize the power line-to-ground impedance to get consistent measurement conditions. The LISN consists of feed-through capacitors and an disk type RF resistor. The measurements confirm the consistency in the impedance value which is maintained at 50 ohms in the frequency range from 1MHz to 500MHz. Thus the newly developed LISN improves consistency of measurement values at all locations, while it was difficult to obtain good correlation before employing the LISN. We feel confident that incorporation of the method discussed here in the pertinent technical standards of EMI measurements, such as CISPR, would lead to a major improvement in getting consistent measurements values.

This paper presents a survey on bicriteria network optimization problems. When there are two conflicting criteria that evaluate the solution, the bicriteria optimization is to find a solution for which these criteria are both acceptably satisfied. Standard approaches to these problems are to combine these two criteria into an aggregated single criterion. Among such problems, this paper first deals with the case in which the aggregated objective function, denoted h(f1(x), f2(x)), is convex in original two objectives f1(x) and f2(x), and, as its special case, it reviews a strongly polynomial algorithm for the bicriteria minimum-cost circulation problem. It then discusses the case in which h is concave and demonstrates that a parametric approach is effective for this case. Several interesting applications in network optimization that belong to this class are also introduced. Finally we deal with the minimum range problems which seek to find a solution such that weights of the components used in the solution are most uniform. We shall present efficient algorithms for solving these problems arised in network optimization.

In general, dynamic channel assignment has a better performance than fixed channel assignment in a cellular mobile communication system. However, it is complex to control the system and a lot of equipments are required in each cell when dynamic channel assignment is applied to a large service area. Therefore, it is effective to limit the size of the service area in order to correct the defects of dynamic channel assignment. So, we propose an application of dynamic channel assignment to a part of a service area when fixed channel assignment is applied to the remaining part of the area. In the system, the efficiency of channel usage in some cells sometimes becomes terribly low. The system has such a problem to be improved. We show that the rearrangement of the channel allocation is effective on the problem.

ATM cross-connect systems, which will be used for provisioning virtual paths (i.e. logical direct connections between exchanges) in future broadband transport networks, simplify network configuration and yield increased routing and capacity allocating flexibility. This paper describes the design of a large capacity ATM cross-connect system that has a multi-stage network structure which requires only one type of switch module. The capacity of the proposed system can be easily increased without service interruptions. To realize cell sequence integrity, a time stamp is added to the self-routing tag. Required time stamp length and efficient module size are discussed.

For the efficient circuit simulation by the direct method, network tearing and latency techniques have been studied. This letter describes a circuit simulator SPLIT with hierarchical decomposition and latency. The block size of the latent subcircuit can be determined dynamically in SPLIT. We apply SPLIT to the MOS circuit simulation and verify its availability.

We present schemes for disseminating information in the n-dimensional hypercube with some faulty nodes/edges. If each processor can send a message to t neighbors at each round, and if the number of faulty nodes/edges is k(kn), then this scheme will broadcast information from any source to all destinations within any consecutive n+[(k+l)/t] rounds. We also discuss the case where the number of faulty nodes is not less than n.