Fujitsu Corporation

Koichi Shirahata

Yasumoto Tomita

Shigeto Suzuki

Takashi Shiraishi

Hiroyuki Fukuda

Takuji Yamamoto

Hiroshi Endo

Katsuhiro Yoda

Makiko Ito

Takashi Arakawa

Michiko Shiraga

Takuya Fukagai

Amir Haderbache

Kyosuke Maeda

An information processing method includes executing a first process of performing a non-linear analysis by iterating a linear analysis, and executing a second process of predicting a residual threshold to be used for determination of convergence of the linear analysis by a prediction model, based on a residual transition and calculation time for each iteration of the linear analysis obtained for each residual threshold using a plurality of experimental values by the first process. The information processing method further includes performing passage of data between the first process and the second process through an inter-process communication using a shared memory set in a memory.

Information processing method, and information processing apparatus

A non-transitory computer-readable storage medium storing a program that causes a processor to execute a process. The process includes selecting a section from among a plurality of sections, performing on an input data group the time-series analysis from a top section to the section when each of a plurality of approximation application degrees is set in the section, based on a result of the performing, for each of the plurality of approximation application degrees is set, calculating an error and an index value for a result of the time-series analysis when an approximation calculation is not applied, the index value indicating a degree of speed-up, storing an approximation application degree in which the error falls within a range, and the index value obtained when the error falls within the range, and executing the time-series analysis on an another input data group by applying a combination of the approximation application degrees.

Non-transitory computer-readable storage medium storing program for performing time-series analysis by calculating approximation calculation application degree, time-series analysis method for performing time-series analysis by calculating approximation calculation application degree, and information processing apparatus for performing time-series analysis by calculating approximation calculation application degree

An information processing method includes: deciding a timing when transfer to a memory is completed in a total time that is a sum of a calculation time at one or plurality of second layers at which calculation is carried out earlier than a first layer regarding a timing when data relating to calculation of the first layer is stored in the memory based on a calculation time estimated in advance regarding each of one layer or a given number of layers in a plurality of layers included in a neural network and a time of transfer of data relating to calculation of each of the one layer or the given number of layers to the memory; and storing the data relating to calculation of the first layer in the memory based on the decided timing in sequentially carrying out calculation of each layer of the neural network.

Information processing apparatus, information processing method, and non-transitory computer-readable storage medium for storing program

A scheduler includes circuitry configured to, based on similarity between execution time and power consumption information of jobs executed in a system, classifies jobs into groups, construct respective time series prediction models for the groups using a power waveform included in each of the groups as teacher data, predict a power waveform at an interval including a first time from each of the constructed time series prediction models, compare a power waveform at an interval including a first time of a job in execution for which power is to be predicted with the predicted power waveform of each of the groups to identify a similar time series prediction model, based on the identified time series prediction model, predict power consumption at a predetermined interval including a second time for the job for which power is to be predicted, and control job execution based on the predicted power consumption.

Job scheduler, job schedule control method, and storage medium

An information processing apparatus includes a processor including a first operation circuit that executes a product-sum operation, a second operation circuit that executes a certain operation, and a resister. The processor executes a first operation including the certain operation in a first layer in a neural network. The processor executes the first operation by a second method of calculating the certain operation by the second operation circuit, in a case where second operation time necessary for the first operation when the certain operation is executed by the second operation circuit is less than memory transfer time. Or the processor executes the first operation by a first method of calculating the certain operation by an approximate calculation by the first operation circuit, in a case where first operation time necessary for the first operation when executed by the first method is less than the memory transfer time.

Information processing apparatus, neural network program, and processing method for neural network

A non-transitory computer-readable recording medium has stored therein a program that causes a computer to execute a process including: dividing a time from a start to an end of execution of a first job into a plurality of time periods, the time being represented in first power consumption information; calculating a cycle of a time-series variation of power consumption within each time period; generating, for each of the plurality of time periods, a prediction model for predicting power consumption of a predetermined future time; acquiring second power consumption information that is an actual measurement value of power consumption according to an elapsed time from a start of execution of a second job that is being executed; and predicting future power consumption of the second job, using the prediction model generated for a time period including the elapsed time from the start of execution of the second job to a present.

Job power predicting method and information processing apparatus

An image recognition apparatus includes a processor including a plurality of arithmetic units; and a memory storing a plurality of data elements, each corresponding to one of candidate regions detected in an image and indicating a location and an evaluation value of the corresponding candidate region. The processor sorts the data elements by calculating in parallel, in reference to evaluation values, indexes each indicating a position of a corresponding one of the data elements in a sorted ordering and transferring in parallel, after the calculation of the indexes, the data elements based on the indexes. The processor selects part of the candidate regions based on the sorted ordering of the data elements.

Image recognition apparatus and image recognition method for sorting candidate regions detected in image

An information processing device includes: a processor that executes a process, the process including: controlling a recognition process that performs, with respect to input neuron data, a hierarchical neural network operation including a weighting operation using a parameter and that holds the neuron data and the parameter of each layer of the neural network in each of memory areas; and performing, in a learning process of learning the parameter of each layer of the neural network from an error that is obtained from a recognition result, regarding the layer in which the neuron data and the parameter are held in the memory areas, control of calculating an error of the neuron data after calculating an error of the parameter.

Controlling memory area for training a neural network

An image recognizing apparatus includes a processor that controls first and second learning processes, the first learning process in second layers including holding, based on a large/small relation between neuron data size and parameter size of the second layer, in a memory area, an error gradient of parameters to be sent to the corresponding layer of the second layers; and the second learning process between first layers including holding, in a memory area of each first layers, an error gradient of parameters to be sent to the corresponding layer of the first layers, which is computed based on the error gradient or an error gradient of a previous layer of the first layers, based on a large/small relation between neuron data size and parameter size of the first layer.

Growing community of inventors

Yokohama, Japan

Koichi Shirahata