JP7161091B2 - Information processing device, information processing method, program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and a program.

In recent years, image classification, big data analysis, and the like have been actively performed using trained models generated by deep learning.

Generating a trained model by deep learning requires learning using a large amount of learning data, requiring enormous computational resources and time. Moreover, since the recognition accuracy changes greatly depending on the learning conditions such as the learning rate and network structure, it is necessary to change the learning conditions little by little and carry out a lot of learning.

Patent Literature 1 describes a technique for generating artificial intelligence through machine learning.

Japanese Patent Application Laid-Open No. 2017-117226

As described above, since deep learning requires enormous computational resources and time, if one learning job is executed by one processing server, the processing server is occupied by one learning job. Therefore, if the processing server cannot execute other jobs or if the processing server stops unexpectedly, there is a problem that the learning up to that point will be wasted.

Therefore, in order to solve the above-mentioned problems, the present invention provides a method for efficiently using computational resources such as processing servers and efficiently generating trained models by dividing and processing learning jobs. intended to

An information processing system of the present invention is an information processing system that generates a trained model using learning data, and is a priority setting receiving means that receives setting of priority of a learning job that is a job related to generation of a trained model. and dividing means for dividing the learning job for which the priority setting is received by the priority setting receiving means into a plurality of divided learning jobs, and executing the divided learning jobs divided by the dividing means according to the priority. and job execution means for generating a learned model.

According to the present invention, by dividing and processing a learning job, it is possible to efficiently use computational resources such as a processing server and efficiently generate a trained model.

A diagram showing an example of the configuration of an information processing system according to the present invention. 1 is a diagram showing an example of the hardware configuration of a client terminal 101, a job management server 102, and a processing server 103 according to the present invention; FIG. Flowchart showing processing contents in the present invention Flowchart showing details of priority determination processing

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a diagram showing an example of the configuration of an information processing system according to the present invention.

The information processing system according to the present invention has a configuration in which a client terminal 101, a job management server 102, and a processing server 103 are communicably connected.

The client terminal 101 is operated by a user to set the priority of a learning job (learning to generate a trained model, including information such as learning conditions (various parameters such as data for learning and network structure)). accept. The received various setting information is transmitted to the job management server. It also has a display unit that displays various data, and has a function of presenting the progress of the learning job received from the job management server, the learning result, and the like to the user.

The job management server 102 controls the execution of the learning job by issuing processing execution instructions to the processing server 103 according to the settings of various learning conditions received by the client terminal 101 . Also, information such as learning results is transmitted to the client terminal 101 .

The processing server 103 executes the learning job according to the learning data and job execution instructions received from the job management server 102 . Although one processing server 103 is shown in FIG. 1, a plurality of processing servers may be provided.

FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing apparatus (client terminal 101, job management server 102, processing server 103) of the present invention.

As shown in FIG. 2, the information processing apparatus includes a CPU (Central Processing Unit) 201, a ROM (Read Only Memory) 202, a RAM (Random Access Memory) 203, an input controller 205, a video controller 206, A memory controller 207 and a communication I/F controller 208 are connected.

The CPU 201 comprehensively controls each device and controller connected to the system bus 204 .

ROM 202 or external memory 211 stores BIOS (Basic Input/Output System) and OS (Operating System), which are control programs executed by CPU 201, computer-readable executable programs for realizing this information processing method, and necessary various types of programs. Holds data (including data tables).

A RAM 203 functions as a main memory, a work area, and the like for the CPU 201 . The CPU 201 loads programs and the like necessary for executing processing from the ROM 202 or the external memory 211 to the RAM 203, and executes the loaded programs to realize various operations.

The input controller 205 controls inputs from input devices such as a keyboard 209 and a pointing device such as a mouse (not shown). When the input device is a touch panel, the user can give various instructions by pressing (touching with a finger or the like) an icon, cursor, or button displayed on the touch panel.

Also, the touch panel may be a touch panel, such as a multi-touch screen, capable of detecting positions touched by multiple fingers.

Video controller 206 controls display on an external output device such as display 210 . The display shall include the display of a notebook computer integrated with the main body. Note that the external output device is not limited to a display, and may be a projector, for example. In addition, an input device is also provided for the device capable of receiving the above-described touch operation.

The video controller 206 can control a video memory (VRAM) for performing display control, and can use part of the RAM 203 as a video memory area, or provide a separate dedicated video memory. It is possible.

A memory controller 207 controls access to the external memory 211 . External memory can be connected to an external storage device (hard disk), flexible disk (FD), or PCMCIA card slot for storing boot programs, various applications, font data, user files, edit files, and various data via an adapter. A compact flash (registered trademark) memory or the like can be used.

A communication I/F controller 209 connects and communicates with an external device via a network, and executes communication control processing in the network. For example, communication using TCP/IP, telephone lines such as ISDN, and communication using 3G lines for mobile phones are possible.

The CPU 201 enables display on the display 210 by, for example, rasterizing an outline font to a display information area in the RAM 203 . The CPU 201 also allows the user to issue instructions using a mouse cursor (not shown) on the display 210 .

Next, the contents of the processing in the present invention will be explained using the flow chart of FIG.

Among the processes shown in the flowchart of FIG. 3, S301 and S312 are processes in which the CPU 201 of the client terminal 101 reads and executes a predetermined control program. Further, S302, S303, S307 to S311 are processes in which the CPU 201 of the job management server 102 reads and executes a predetermined control program. Further, S304 to S306 are processes in which the CPU 201 of the processing server 103 reads and executes a predetermined control program.

In step S<b>301 , the client terminal 101 receives setting of the priority of the learning job from the user. Then, the received information such as the priority is transmitted to the job management server 102 .

In step S302, the job management server 102 divides the learning job to be processed upon receiving the information such as the priority of the job transmitted from the client terminal 101. FIG.

In step S303, a priority is set for the job (divided learning job) divided in step S302, and the processing server 103 is instructed to execute the divided learning job.

In step S<b>304 , the processing server 103 acquires a high-priority divided learning job in accordance with an instruction from the job management server 102 .

In step S305, the divided learning job acquired in step S304 is executed.

In step S<b>306 , the result of the divided learning job executed in step S<b>305 is transmitted to the job management server 102 .

In step S307, when receiving the result of the divided learning job from the processing server 103, the job management server 102 determines whether the divided learning job is the last divided learning job (for example, the learning job A is divided into five of A1 to A5). is the fifth divided learning job in the case of

If it is the last divided learning job (step S307: YES), the process proceeds to step S311, and the learning job is brought to an end state. Then, the learning result of the learning job is transmitted to the client terminal 101, and the client terminal 101 receives the transmitted learning result (S312).

If it is not the last divided learning job (step S307: NO), the process proceeds to step S308.

In step S308, based on the result of the divided learning job by the processing server 103, the timing for the next division of the learning job is determined. Specifically, when it is considered that over-learning has occurred from the learning result, it is controlled to be finely divided. By doing so, it is possible to reduce unnecessary over-learning. Further, by finely dividing and executing the processing of step S309, which will be described later, computational resources can be used efficiently.

It is also possible to prevent overlearning by changing the parameters to lower the learning rate and restarting learning from the divided job immediately before overlearning occurs.

In step S<b>309 , the priority of the divided learning job is controlled based on the result of the divided learning job by the processing server 103 .

Details of the processing in step S309 will be described with reference to the flowchart of FIG.

In step S401, the job management server 102 determines whether the loss function of the result of the divided learning job by the processing server 103 has diverged. That is, it is determined whether or not the expected result is not obtained and the learning is failed.

When the loss function diverges (when it is determined that the learning has failed) (step S401: YES), the setting of the learning conditions is inappropriate, and the priority is given to continuing the processing of the learning job. Therefore, the priority of the learning job is lowered (S404).

If the loss function has not diverged (step S401: NO), the process proceeds to step S402.

In step S402, it is determined whether the learning is already sufficient (if the expected results are obtained, it is determined that the learning is sufficient).

If the learning is already sufficient (step S402: YES), it is considered that the current learning progress is sufficient and the priority of continuing the processing of the learning job is considered to be low, so the process proceeds to step S405. Then, the priority of the learning job is lowered (S405).

If the learning is still insufficient (step S402: NO), the process proceeds to step S403.

In step S403, it is determined whether the learning result is better than the learning job with other learning conditions. If the results are better than the jobs with other learning conditions (step S403: YES), it is likely that the expected results will be obtained by continuing the learning job, so the learning job is prioritized. Increase the degree (S406).

If the result is worse than the learning job with other learning conditions (step S403: NO), it is considered unlikely that the expected result will be obtained even if the learning job is continuously executed. Lower the priority (S407).

The details of the processing in step S309 in FIG. 3 have been described above.

Then, in step S310, the job is divided into divided learning jobs such that learning is restarted from the intermediate results, and the process returns to step S303.
In step S303, the priority determined in step S309 is set for the divided learning job, and the processing server 103 is instructed to execute the divided learning job.

In this way, by dividing the learning jobs and executing the learning jobs according to their priority, even if the processing server stops while the learning jobs are being executed, the learning results executed up to that point (processing Therefore, the results of the divided learning jobs that have been divided are not wasted.

Further, even when a learning job with a high priority is newly registered, the processing can be switched to a learning job with a high priority at the timing when the divided learning job ends.

Also, in an environment that uses both an internal server that can be used freely and an external server that is billed according to usage, it is possible to switch to executing processing on the internal server at the timing when the divided learning job is completed, resulting in cost reduction. is also effective.

The present invention can be embodied as, for example, a system, device, method, program, recording medium, or the like. Specifically, it may be applied to a system composed of a plurality of devices, or may be applied to an apparatus composed of a single device.

Further, the program in the present invention is a program that allows a computer to execute the processing methods of the flow charts shown in FIGS. is stored. The program according to the present invention may be a program for each processing method of each device shown in FIGS.

As described above, a recording medium recording a program for realizing the functions of the above-described embodiments is supplied to a system or device, and the computer (or CPU or MPU) of the system or device reads the program stored in the recording medium. Needless to say, the object of the present invention can also be achieved by reading and executing.

In this case, the program itself read from the recording medium implements the novel functions of the present invention, and the recording medium recording the program constitutes the present invention.

Examples of recording media for supplying programs include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, CD-Rs, DVD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, EEPROMs, silicon A disk or the like can be used.

Further, by executing the program read by the computer, not only the functions of the above-described embodiments are realized, but also based on the instructions of the program, the OS (operating system) and the like running on the computer are actually executed. Needless to say, a case where part or all of the processing is performed and the functions of the above-described embodiments are realized by the processing are included.

Furthermore, after the program read from the recording medium is written in the memory provided in the function expansion board inserted into the computer or the function expansion unit connected to the computer, the function expansion board is read according to the instruction of the program code. It goes without saying that a case where a CPU or the like provided in a function expansion unit performs part or all of the actual processing and the functions of the above-described embodiments are realized by the processing.

Moreover, the present invention may be applied to a system composed of a plurality of devices or to an apparatus composed of a single device. Further, it goes without saying that the present invention can be applied to the case where it is achieved by supplying a program to a system or device. In this case, by loading a recording medium storing a program for achieving the present invention into the system or device, the system or device can enjoy the effects of the present invention.

Furthermore, by downloading and reading out the program for achieving the present invention from a server, database, etc. on the network using a communication program, the system or device can enjoy the effects of the present invention. It should be noted that all configurations obtained by combining each of the above-described embodiments and modifications thereof are also included in the present invention.

101 client terminal 102 job management server 103 processing server

Claims (6)

An information processing system that generates a trained model using learning data,
a priority setting reception means for receiving a priority setting for execution of a learning job, which is a job for generating a trained model;
a dividing unit that divides the learning job for which the priority setting is received by the priority setting receiving unit into a plurality of divided learning jobs;
a job executing means for generating a trained model by executing the divided learning jobs divided by the dividing means according to the priority;
priority control means for changing the priority of execution of the learning job related to the divided learning job received by the priority setting receiving means according to the result of execution of the divided learning job by the job executing means;
An information processing system comprising: When it is determined that learning has failed as a result of execution of the divided learning job, the priority control means lowers the priority of execution of the learning job related to the divided learning job. 2. The information processing system according to claim 1. When it is determined that learning has been sufficiently completed as a result of execution of the divided learning job, the priority control means lowers the priority of execution of the learning job related to the divided learning job. 3. The information processing system according to claim 1 or 2, characterized by: 2. When it is determined that over-learning has occurred as a result of execution of the divided learning job, the dividing unit further divides the learning job related to the divided learning job. 4. The information processing system according to any one of items 1 to 3. An information processing method in an information processing system that generates a trained model using learning data,
a priority setting receiving step of receiving a priority setting related to execution of a learning job, which is a job related to generation of a trained model;
a dividing step of dividing the learning job for which the priority setting is received by the priority setting receiving step into a plurality of divided learning jobs;
a job execution step of generating a learned model by executing the divided learning jobs divided by the dividing step according to the priority;
a priority control step of changing the priority of execution of the learning job related to the divided learning job received by the priority setting receiving step according to the result of execution of the divided learning job by the job executing step;
An information processing method comprising: A program for causing a computer to function as each means of the information processing system according to any one of claims 1 to 4.

Images