# US 2022/0277222 A1 · Storage medium, machine learning method, and information processing device

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0277222 A1
- Document ID: US20220277222A1
- Status: Published application
- Kind: A1
- Application number: 17/521,914
- Filed: 2021-11-09
- Published: 2022-09-01
- Earliest claimed: 2021-02-26
- Assignees: Fujitsu Limited
- Applicants: Fujitsu Limited
- Inventors: Koichi SHIRAHATA; Masahiro Miwa
- Web page: https://patentgrep.com/patent/US20220277222A1

## CPC

- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 20/00: Machine learning
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 3/098: Distributed learning, e.g. federated learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A non-transitory computer-readable storage medium storing a machine learning program that causes at least one computer to execute a process, the process includes, training a machine learning model by using a backpropagation process; skipping reading a first mini-batch in a first epoch among a plurality of mini-batches that are created by dividing training data.

## Claims

### Claim 1

A non-transitory computer-readable storage medium storing a machine learning program that causes at least one computer to execute a process, the process comprising:
training a machine learning model by using a backpropagation process; and
skipping reading a first mini-batch in a first epoch among a plurality of mini-batches that are created by dividing training data.

### Claim 2 (depends on 1)

The non-transitory computer-readable storage medium according to claim 1, wherein the process further comprising
skipping reading a second mini-batch in a second epoch.

### Claim 3 (depends on 1)

The non-transitory computer-readable storage medium according to claim 1, wherein the process comprising
when the training includes using a plurality of processors to train parallelly, causing the plurality of processors to skip reading the first mini-batch at a same timing.

### Claim 4 (depends on 3)

The non-transitory computer-readable storage medium according to claim 3, wherein the process further comprising
when the training includes using a plurality of processors to train parallelly, causing one of the plurality of processors with a lower processing speed to skip reading a greater number of mini-batches than others of the plurality of processors according to respective processing speeds of the plurality of processors.

### Claim 5 (depends on 2)

The non-transitory computer-readable storage medium according to claim 2, wherein the process further comprising:
acquiring a degree of training influence on the machine learning for each of the mini-batches in the first epoch by calculating; and
preferentially skipping reading a mini-batch whose degree of training influence is low.

### Claim 6 (depends on 3)

The non-transitory computer-readable storage medium according to claim 3, wherein the process further comprising:
acquiring a difference between a first training efficiency when training the machine learning model by using all of the plurality of processors and a second training efficiency when training the machine learning model by using other ones of the processors excluded one processor from the plurality of processors; and
skipping reading mini-batches by the one particular processor when the difference falls below a threshold value.

### Claim 7

A machine learning method for a computer to execute a process comprising:
training a machine learning model by using a backpropagation process; and
skipping reading a first mini-batch in a first epoch among a plurality of mini-batches that are created by dividing training data.

### Claim 8 (depends on 7)

The machine learning method according to claim 7, wherein the process further comprising
skipping reading a second mini-batch in a second epoch.

### Claim 9 (depends on 7)

The machine learning method according to claim 7, wherein the process comprising
when the training includes using a plurality of processors to train parallelly, causing the plurality of processors to skip reading the first mini-batch at a same timing.

### Claim 10 (depends on 9)

The machine learning method according to claim 9, wherein the process further comprising
when the training includes using a plurality of processors to train parallelly, causing one of the plurality of processors with a lower processing speed to skip reading a greater number of mini-batches than others of the plurality of processors according to respective processing speeds of the plurality of processors.

### Claim 11 (depends on 9)

The machine learning method according to claim 9, wherein the process further comprising:
acquiring a degree of training influence on the machine learning for each of the mini-batches in the first epoch by calculating; and
preferentially skipping reading a mini-batch whose degree of training influence is low.

### Claim 12 (depends on 10)

The machine learning method according to claim 10, wherein the process further comprising:
acquiring a difference between a first training efficiency when training the machine learning model by using all of the plurality of processors and a second training efficiency when training the machine learning model by using other ones of the processors excluded one processor from the plurality of processors; and
skipping reading mini-batches by the one particular processor when the difference falls below a threshold value.

### Claim 13

An information processing device, comprising:
a memory; and
a processor coupled to the memory and configured to:
train a machine learning model by using a backpropagation process, and
skip reading a first mini-batch in a first epoch among a plurality of mini-batches that are created by dividing training data.
