R-MAT (for Recursive MATrix) is a simple, widely used model for generating graphs with a power law degree distribution, a small diameter, and communitys structure. It is particularly attractive for generating very large graphs because edges can be generated independently by an arbitrary number of processors. However, current R-MAT generators need time logarithmic in the number of nodes for generating an edge— constant time for generating one bit at a time for node IDs of the connected nodes. We achieve constant time per edge by precomputing pieces of node IDs of logarithmic length. Using an alias table data structure, these pieces can then be sampled in constant time. This simple technique leads to practical improvements by an order of magnitude. This further pushes the limits of attainable graph size and makes generation overhead negligible in most situations.

A parallel-processing system for locating parts and controlling an industrial robot is proposed. The system employs Transputers and Occam to achieve parallelism. In conjunction with a novel vibratory sensor, the system enables a robot to determine the exact location of parts which have been picked up from a semi-ordered work place. A new algorithm for obtaining the coordinates of the parts using the sensed vibration and deflection signals is described. The algorithm dispenses with the lengthy and complex equation-solving procedures previously required. Instead, it only involves looking up a data table and performing simple two-dimensional interpolation calculations. The design of the algorithm to ensure efficient parallel operation is described. Experimental results showing the successful implementation of the algorithm on the proposed system are presented.

We study the asymptotic behavior of a new particle filter approach for the estimation of hidden Markov models. In particular, we develop an algorithm where the latent-state sequence is segmented into multiple shorter portions, with an estimation technique based upon a separate particle filter in each portion. The partitioning facilitates the use of parallel processing, which reduces the wall-clock computational time. Based upon this approach, we introduce new estimators of the latent states and likelihood which have similar or better variance properties compared to estimators derived from standard particle filters. We show that the likelihood function estimator is unbiased, and show asymptotic normality of the underlying estimators.

This paper presents a robust reliability analysis method for systems of multimodular redundant (MMR) controllers using the method of partitioning and parallel processing of a Markov chain (PPMC). A Markov chain is formulated to represent the N distinct states of the MMR controllers. Such a Markov chain has N2 directed edges, and each edge corresponds to a transition probability between a pair of start and end states. Because N can be easily increased substantially, the system reliability analysis may require large computational resources, such as the central processing unit usage and memory occupation. By the PPMC, a Markov chain's transition probability matrix can be partitioned and reordered, such that the system reliability can be evaluated through only the diagonal submatrices of the transition probability matrix. In addition, calculations regarding the submatrices are independent of each other and thus can be conducted in parallel to assure the efficiency. The simulation results show that, compared with the sequential method applied to an intact Markov chain, the proposed PPMC can improve the performance and produce allowable accuracy for the reliability analysis on large-scale systems of MMR controllers.

Two experiments were conducted to test the hypothesis that the parallel processing of the target language (TL) during source language (SL) comprehension in interpreting may be influenced by two factors: (i) link strength from SL to TL, and (ii) the interpreter's cognitive resources supplement to TL processing during SL comprehension. The influence of the first factor was supported by the contrasting performance on bidirectional SL and TL interpreting tasks by unbalanced bilingual student interpreters, and the second factor was supported by the contrasting performance between participants’ two developmental stages in interpreting. Implications are discussed.

Prevailing engineering-inspired theories of motor control based on sequential/algorithmic or motor-programming models are difficult to reconcile with what is known about the anatomy and physiology of the motor areas. This is partly because of certain problems with the theories themselves and partly because of features of the cortical and basal ganglionic motor circuits that seem ill-suited for most engineering analyses of motor control. Recent developments in computational neuroscience offer more realistic, that is, connectionist, models of motor processing. The distributed, highly parallel, and nonalgorithmic processes in these models are inherently self-organizing and hence more plausible biologically than their more traditional algorithmic or motor-programming counterparts. The newer models also have the potential to explain some of the unique features of natural, brain-based motor behavior and to avoid some of the computational dilemmas associated with engineering approaches.

The cerebral cortex is a rich and diverse structure that is the basis of intelligent behavior. One of the deepest mysteries of the function of cortex is that neural processing times are only about one hundred times as fast as the fastest response times for complex behavior. At the very least, this would seem to indicate that the cortex does massive amounts of parallel computation.

This paper explores the hypothesis that an important part of the cortex can be modeled as a connectionist computer that is especially suited for parallel problem solving. The connectionist computer uses a special representation, termed value unit encoding, that represents small subsets of parameters in a way that allows parallel access to many different parameter values. This computer can be thought of as computing hierarchies of sensorimotor invariants. The neural substrate can be interpreted as a commitment to data structures and algorithms that compute invariants fast enough to explain the behavioral response times. A detailed consideration of this model has several implications for the underlying anatomy and physiology.

Male Cebus monkeys are all dichromats, but about two thirds of the females are trichromats. M and P retinal ganglion cells were studied in the male Cebus monkey to investigate the relationship of their morphology to retinal eccentricity. Retinal ganglion cells were retrogradely labeled after optic nerve deposits of biocytin to reveal their entire dendritic tree. Cebus M and P ganglion cell morphology revealed by biocytin retrograde filling is similar to that described for macaque and human M and P ganglion cells obtained by in vitro intracellular injection of HRP and neurobiotin. We measured 264 and 441 M and P ganglion cells, respectively. M ganglion cells have larger dendritic field and cell body size than P ganglion cells at any comparable temporal or nasal eccentricity. Dendritic trees of both M and P ganglion cells are smaller in the nasal than in the temporal region at eccentricities greater than 5 mm and 2 mm for M and P ganglion cells, respectively. The depth of terminal dendrites allows identification of both inner and outer subclasses of M and P ganglion cells. The difference in dendritic tree size between inner and outer cells is small or absent. Comparison between Cebus and Macaca shows that M and P ganglion cells have similar sizes in the central retinal region. The results support the view that M and P pathways are similarly organized in diurnal dichromat and trichromat primates.

The vertebrate retina processes visual information in parallel neural pathways known as the ON and OFF pathways. These pathways encode increments and decrements of light independently as excitatory responses. We examined the photopic spectral response of ON and OFF mechanisms in goldfish by measuring the sensitivity of optic nerve responses to the onset and termination of stimuli of various wavelengths. Using various adapting backgrounds, we found that the ON and OFF responses have different spectral sensitivities. The weighting of the cone inputs to the responses was estimated by an algebraic summation model. This model suggests that for the ON response, input from S-cones is stronger and more independent than for the OFF response, and M- and L-cones show stronger antagonism in the ON response than in the OFF response. The OFF response probably receives input from all cone types, but spectral antagonism is weak and its dominant input is from L-cones.

The first (V-I) and second (V-II) visual areas of primates contain three types of anatomical segregations of neurons as parts of hypothesized “P-B” or “color”, “P-I” or “form,” and “M” or “motion” processing channels. These channels remain distinct in relays of P-B and P-I information to the inferior temporal lobe via V-II and dorsolateral visual cortex for object recognition, and “M” information to posterior parietal cortex via the middle temporal visual area (MT) for visual tracking and attention. The present anatomical experiments demonstrate another channel where “P-B” modules in V-I and “P-B” and “M” modules in V-II merge in the projections to the dorsomedial visual area (DM), which relays to MT and posterior parietal cortex. This integrative area may function in unifying our perception of the visual world, and may allow “color” as well as “motion” to play a role in visual tracking and attention.

The accurate modelling of robot dynamics is essential for the design of model-based robot controllers. However, dynamic models have very complicated features which can be attributed to several reasons. For example, the continuously-varying arm configuration, uncertain effects of load handling on the dynamic stability of the arm, and the high degree of non-linearity and coupling exhibited between the different links. Hence, the accurate modelling of these effects will play an important role in the design of robust controllers. Towards this end, an efficient and fast method for the on-line tuning of robot dynamic parameters must be devised. This work proposes to solve this problem as follows. First, a simplified dynamic model of the robot is developed. The model allows for direct and straightforward extraction and regrouping of dynamic parameters. The resulting dynamic parameters are formulated as a regression equation which is linear in the dynamic parameters. Finally, the algorithm is executed using a Transputer development system to speed up the computation and meet real-time constraints. The efficiency of the approach is demonstrated by a case study.

We study a fork-join processing network in which jobs arrive according to a Poisson process and each job splits into m tasks, which are simultaneously assigned to m nodes that operate like M/M/s queueing systems. When all of its tasks are finished, the job is completed. The main result is a closed-form formula for approximating the distribution of the network's response time (the time to complete a job) in equilibrium. We also present an analogous approximation for the distribution of the equilibrium queue length (the number of jobs in the system), when each node has one server. Kolmogorov-Smirnov statistical tests show that these formulae are good fits for the distributions obtained from simulations.

In this paper we consider the problem of scheduling precedence task graphs inparallel processing where there can be disturbances in computation andcommunication times. Such a phenomenon often occurs in practice, due to ourinability to exactly predict the time because of system intrusion like cache miss and packet transmission time in mediums like ethernet etc. We propose a method based on the addition of some extra edges to protect the initial scheduling from performing badly due to such changes and provide an upper bound on theperformance guarantee for the scheduling algorithms. Moreover, this construction guarantees a result at least as goodas the result obtained for the initial static scheduling.We also show that this construction is a minimal set in context of partially on-line scheduling.

In this work scheduling multiprocessor tasks on two parallel identical processors is considered.Multiprocessor tasks can be executed by more than one processorat the same moment of time.We analyze scheduling unit execution time and preemptable tasks to minimize schedule length and maximum lateness.Cases with ready times, due-dates and precedence constraintsare discussed.

In this paper, we present a new linear time algorithm forscheduling UECT (Unit Execution and Communication Time) trees on twoidentical processors. The chosen criterion is the makespan. The used strategy is based on clustering of tasks. We show that this algorithm builds optimal schedules. Some extensions are discussed for non UECT tasks.

We give a new formulation for the problem of task scheduling intounrelated processors under precedence constraints. This formulation has a polynomial number of variables and does not require that theprocessing times be integer valued.

This paper presents an investigation into the utilisation of sequential and parallel processing techniques for the real-time simulation of a flexible manipulator system. A finite dimensional simulation of the system is developed using a finite difference approximation to the governing dynamic equation of the manipulator. The developed algorithm is implemented on a number of uni-processor and multi-processor, homogeneous and heterogeneous, parallel architectures. A comparison of the results of these implementations is made and discussed, on the basis of real-time processing requirements in the simulation and control of flexible manipulator systems.

We measured the photopic spectral sensitivity of multiunit activity in the torus longitudinalis and optic tectum of goldfish. Since negative contrast stimuli are most effective for exciting torus longitudinalis, spiking activity was evoked by the shadow of a disc moving through a monochromatic light beam projected upon a screen. The amount of activity evoked in torus longitudinalis generally increased with the monochromatic stimulus radiance at the same rate for all wavelengths, indicating a univariant response. Spiking activity in tectum, however, increased at different rates across the spectrum, indicating color-dependent responses. The action spectra for torus longitudinalis were all similar and relatively flat as expected of a homogeneous, broad-band luminance processing system, and about 1 log unit more sensitive than the tectal action spectra. The latter generally displayed sharp peaks and dips in sensitivity indicative of opponent processing.

We obtain a limit theorem for the joint distribution of the maximum value and sample mean of a random length sequence of independent and identically distributed random variables. This extends a previous bivariate convergence result for fixed length sequences and incidentally yields a new proof of Berman's classical limit theorem for the maximum value of a random number of random variables. Our approach uses a property of record time sequences and leads to probabilistically intuitive proofs. We also consider the partition of a finite interval into a random number of subintervals by the points of a non-delayed renewal process. Using the bivariate convergence result for random length sequences, we establish a limit theorem for the joint distribution of the number and maximum length of the subintervals as the interval length becomes large. This leads to limiting results for the ratio of the maximum to the mean subinterval length. Such results are of interest in connection with a simple model of parallel processing.