WO2023127062A1 - データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム - Google Patents

データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム Download PDF

Info

Publication number
WO2023127062A1
WO2023127062A1 PCT/JP2021/048702 JP2021048702W WO2023127062A1 WO 2023127062 A1 WO2023127062 A1 WO 2023127062A1 JP 2021048702 W JP2021048702 W JP 2021048702W WO 2023127062 A1 WO2023127062 A1 WO 2023127062A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
pseudo
ood
machine learning
updating
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/JP2021/048702
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
智尊 江田
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujitsu Ltd
Original Assignee
Fujitsu Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fujitsu Ltd filed Critical Fujitsu Ltd
Priority to PCT/JP2021/048702 priority Critical patent/WO2023127062A1/ja
Priority to EP21969944.4A priority patent/EP4459514A4/en
Priority to JP2023570547A priority patent/JPWO2023127062A1/ja
Publication of WO2023127062A1 publication Critical patent/WO2023127062A1/ja
Priority to US18/739,788 priority patent/US20240330684A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Definitions

  • the present invention aims to enable efficient generation of out-of-distribution data.
  • the OOD data is input to a discriminator (not shown).
  • the discriminator discriminates whether the input data is IND or OOD, and may output a value representing the closeness to the IND data.
  • the discriminator may be denoted by symbol D.
  • data to be processed is image data, and the OOD data generation unit 100 generates image data as the OOD data.
  • ⁇ is a parameter (hyperparameter) for adjusting the trade-off between L c (x, t) and L d (x), and ⁇ [0, 1].
  • the OOD data generation unit 100 adjusts the parameter ⁇ to optimize the trade-off between L c (x, t) and L d (x), thereby generating OOD samples that are difficult for classifiers to identify.
  • FIG. 2 is a diagram showing a method of updating OOD samples x ⁇ in the information processing device 1 as an example of the first embodiment.
  • step S3 the discriminator update unit 103 updates the discriminator C so that the OOD candidate data updated in step S2 is recognized as OOD. As a result, the discriminator C does not discriminate the surroundings of this OOD candidate data with a high degree of certainty.
  • FIG. 4 shows a program-like process of generating the OOD sample X ⁇ by optimization . to generate
  • the OOD data candidate update unit 102 updates the OOD data candidates generated by the OOD data candidate generation unit 101 so that the loss for the class is minimized. update to
  • OOD data for training with high coverage can be generated for the class determination model.
  • OOD which is completely different from IND, which discriminators are not good at, but which discriminates with a high degree of certainty.
  • FIG. 7 is a diagram schematically showing an OOD generation model provided in the information processing device 1 as an example of the second embodiment.
  • the OOD generation model 104 is a neural network having weights W, receives data z, and outputs OOD data G(z).
  • the neural network may be a hardware circuit, or a software virtual network that connects layers virtually constructed on a computer program by a processor 11 (see FIG. 10), which will be described later.
  • the machine learning model may include the OOD generation model 104 and the discriminator 105 .
  • the OOD generation model 104 is trained (machine learning) to generate samples that have a high degree of confidence in the discriminator 105 and are different from IND.
  • the training data generation unit 201 may randomly generate OOD original data (pseudo data) and dummy labels for the OOD original data. You may represent OOD original data using code
  • the parameter setting unit 202 updates the parameters of the neural network in the direction of decreasing the loss function that defines the error between the inference result of the OOD generation model 104 for the training data and the correct data using, for example, the gradient descent method.
  • Parameter optimization may be performed by
  • the OOD generative model 104 generator G
  • FIG. 9 is a diagram illustrating an algorithm for optimization processing of the OOD generation model 104 in the information processing device 1 as an example of the second embodiment.
  • FIG. 9 shows the process of optimizing the OOD generation model 104 in a program-like manner.
  • the generated OOD data is added to the training data Dtr , and the classifier C is updated using these data.
  • the updating of this classifier C is realized by regularizing and learning the OOD data so as to output a uniform output (see reference P09).
  • a monitor 14a is connected to the graphics processing device 14.
  • the graphics processing unit 14 displays an image on the screen of the monitor 14a in accordance with instructions from the processor 11.
  • FIG. Examples of the monitor 14a include a display device using a CRT (Cathode Ray Tube), a liquid crystal display device, and the like.
  • the optical drive device 16 uses laser light or the like to read data recorded on the optical disk 16a.
  • the optical disc 16a is a portable, non-temporary recording medium on which data is recorded so as to be readable by light reflection.
  • the optical disk 16a includes DVD (Digital Versatile Disc), DVD-RAM, CD-ROM (Compact Disc Read Only Memory), CD-R (Recordable)/RW (ReWritable), and the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Machine Translation (AREA)
PCT/JP2021/048702 2021-12-27 2021-12-27 データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム Ceased WO2023127062A1 (ja)

Priority Applications (4)

Application Number Priority Date Filing Date Title
PCT/JP2021/048702 WO2023127062A1 (ja) 2021-12-27 2021-12-27 データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム
EP21969944.4A EP4459514A4 (en) 2021-12-27 2021-12-27 DATA GENERATION METHOD, MACHINE LEARNING METHOD, INFORMATION PROCESSING APPARATUS, DATA GENERATION PROGRAM AND MACHINE LEARNING PROGRAM
JP2023570547A JPWO2023127062A1 (https=) 2021-12-27 2021-12-27
US18/739,788 US20240330684A1 (en) 2021-12-27 2024-06-11 Data generation method, machine learning method, information processing apparatus, non-transitory computer-readable recording medium storing data generation program, and non-transitory computer-readable recording medium storing machine learning program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2021/048702 WO2023127062A1 (ja) 2021-12-27 2021-12-27 データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/739,788 Continuation US20240330684A1 (en) 2021-12-27 2024-06-11 Data generation method, machine learning method, information processing apparatus, non-transitory computer-readable recording medium storing data generation program, and non-transitory computer-readable recording medium storing machine learning program

Publications (1)

Publication Number Publication Date
WO2023127062A1 true WO2023127062A1 (ja) 2023-07-06

Family

ID=86998370

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2021/048702 Ceased WO2023127062A1 (ja) 2021-12-27 2021-12-27 データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム

Country Status (4)

Country Link
US (1) US20240330684A1 (https=)
EP (1) EP4459514A4 (https=)
JP (1) JPWO2023127062A1 (https=)
WO (1) WO2023127062A1 (https=)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117995173A (zh) * 2024-01-31 2024-05-07 三六零数字安全科技集团有限公司 一种语言模型生成方法、装置、存储介质及电子设备

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2020123830A (ja) 2019-01-30 2020-08-13 京セラドキュメントソリューションズ株式会社 画像処理装置、画像読取装置、画像形成装置、画像処理方法及び画像処理プログラム
US20210182731A1 (en) 2019-12-13 2021-06-17 Robert Bosch Gmbh Reciprocating generative models

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2020123830A (ja) 2019-01-30 2020-08-13 京セラドキュメントソリューションズ株式会社 画像処理装置、画像読取装置、画像形成装置、画像処理方法及び画像処理プログラム
US20210182731A1 (en) 2019-12-13 2021-06-17 Robert Bosch Gmbh Reciprocating generative models

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
See also references of EP4459514A4
TANG KEKE; MIAO DINGRUIBO; PENG WEILONG; WU JIANPENG; SHI YAWEN; GU ZHAOQUAN; TIAN ZHIHONG; WANG WENPING: "CODEs: Chamfer Out-of-Distribution Examples against Overconfidence Issue", 2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), IEEE, 10 October 2021 (2021-10-10), pages 1133 - 1142, XP034093906, DOI: 10.1109/ICCV48922.2021.00119 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117995173A (zh) * 2024-01-31 2024-05-07 三六零数字安全科技集团有限公司 一种语言模型生成方法、装置、存储介质及电子设备

Also Published As

Publication number Publication date
JPWO2023127062A1 (https=) 2023-07-06
EP4459514A4 (en) 2025-02-19
US20240330684A1 (en) 2024-10-03
EP4459514A1 (en) 2024-11-06

Similar Documents

Publication Publication Date Title
Sohn et al. Arachne: Search-based repair of deep neural networks
JP6781415B2 (ja) ニューラルネットワーク学習装置、方法、プログラム、およびパターン認識装置
Lad et al. Malware classification with improved convolutional neural network model
Liu et al. Security analysis and enhancement of model compressed deep learning systems under adversarial attacks
KR20210149530A (ko) 이미지 분류 모델 학습 방법 및 이를 수행하기 위한 장치
JP2008009893A (ja) パラメータ学習方法及びその装置、パターン識別方法及びその装置、プログラム
CN115630303B (zh) 一种基于类激活映射的对抗样本可视化解释方法
KR102173942B1 (ko) 다른 시점의 영상을 이용한 객체 검출을 위한 장치 및 이를 위한 방법
KR102615055B1 (ko) 적대적 이미지 복원 시스템 및 적대적 이미지 복원 방법
US11715197B2 (en) Image segmentation method and device
García-Manso et al. Study of the effect of breast tissue density on detection of masses in mammograms
KR20240121054A (ko) 결함 검출 모델을 학습시키는 학습 방법 및 학습 장치
US20210342430A1 (en) Identity verification using task-based behavioral biometrics
Vashishtha et al. An Ensemble approach for advance malware memory analysis using Image classification techniques
US12456284B2 (en) Device and method for classifying images and accessing the robustness of the classification
KR20200052411A (ko) 영상 분류 장치 및 방법
WO2023127062A1 (ja) データ生成方法，機械学習方法，情報処理装置，データ生成プログラムおよび機械学習プログラム
CN119026127B (zh) 基于多层次特征融合的恶意代码检测方法、系统及设备
EP4036776B1 (en) Information processing program, information processing method, and information processing apparatus
KR102902700B1 (ko) 적대적 이미지 탐지 시스템 및 적대적 이미지 탐지 방법
JP2023183079A (ja) 訓練データ生成プログラム，訓練データ生成方法および情報処理装置
JP7367859B2 (ja) 学習装置、学習済みモデル生成方法、分類装置、分類方法、及びプログラム
Mercaldo et al. Detecting Retinal Fundus Image Synthesis by Means of Generative Adversarial Network.
Hashmi et al. Enhanced Wasserstein Generative Adversarial Network (EWGAN) to Oversample Imbalanced Datasets
Sun et al. CAPTCHA recognition based on Kohonen maps

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21969944

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2023570547

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2021969944

Country of ref document: EP

Effective date: 20240729