Anomalous/Exceptional Pattern Mining with Weak Label Information from Stream Data

• Project/Area Number: 18H03290
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Kyushu University
• Principal Investigator: Suzuki Einoshin 九州大学, システム情報科学研究院, 教授 (10251638)
• Co-Investigator(Kenkyū-buntansha): 安藤 晋 東京理科大学, 経営学部ビジネスエコノミクス学科, 准教授 (70401685)
• Project Period (FY): 2018-04-01 – 2021-03-31
• Project Status: Completed (Fiscal Year 2020)
• Budget Amount: ¥17,420,000 (Direct Cost: ¥13,400,000、Indirect Cost: ¥4,020,000); Fiscal Year 2020: ¥5,200,000 (Direct Cost: ¥4,000,000、Indirect Cost: ¥1,200,000); Fiscal Year 2019: ¥4,940,000 (Direct Cost: ¥3,800,000、Indirect Cost: ¥1,140,000); Fiscal Year 2018: ¥7,280,000 (Direct Cost: ¥5,600,000、Indirect Cost: ¥1,680,000)
• Keywords: 異常・例外発見 / ストリームデータマイニング / 弱ラベル学習 / イメージキャプショニング / ファスト＆スロー思考 / 単語埋め込み / パターンマイニング

Outline of Final Research Achievements

We made achievements in various problems to discover anomalies and exceptionalities by assigning weak label information from dynamic data. Especially, we have proposed two methods which rapidly and accurately detect anomalous image regions and their combinations by assigning captions as weak labels with a deep neural network for rapidly observed image sequence data and confirmed their effectiveness by various means including experiments using an autonomous mobile robot. These methods model normal data with fast clustering in the training phase and then detect anomalous data which differ from normal data in the test phase. The first method won the Best Poster Prize in an international conference. The second method mimics fast & slow thinking done by human beings and detects more complex exceptionalities rapidly.

Academic Significance and Societal Importance of the Research Achievements

訓練フェーズで正常データをモデリングし，テストフェーズで正常データとは異なる異常データを検知する1クラス異常検知問題は，その実用的価値の高さと学術的困難さから，データマイニングと機械学習における重要問題である．本研究成果はこの問題に対し，深層学習に基づいて自動特定した重要な画像領域群に自動付与された説明文を疑似教師信号の一種である弱ラベルとして有効活用する方式を初めて提案した手法であり，たとえば外見が大きく異なる2人が共に女性であるという手がかりを活かせる．この種の説明文はしばしば不正確であり，高速に観測される動的データには実時間処理が必須だが，いずれの問題も高いレベルで解決している．

Research Products

• [Journal Article] Joint Optimization of Neural Collaborative Filtering and Variational Auto-Encoder for Hybrid Recommendation (2021)
  • Author(s): 園田亮介，鈴木英之進
  • Journal Title: 電子情報通信学会論文誌D 情報・システム Volume: J104-D Issue: 2 Pages: 119-129
  • DOI: 10.14923/transinfj.2020JDP7017
  • ISSN: 1880-4535, 1881-0225
  • Year and Date: 2021-02-01
  • Peer Reviewed / Open Access
• [Journal Article] Topic Modeling for Sequential Documents Based on Hybrid Inter-Document Topic Dependency (2021)
  • Author(s): Wenbo Li, Tetsu Matsukawa, Hiroto Saigo, Einoshin Suzuki
  • Journal Title: Journal of Intelligent Information Systems Volume: 1 Issue: 3 Pages: 435-458
  • DOI: 10.1007/s10844-020-00635-4
  • Peer Reviewed / Open Access
• [Journal Article] Adaptive and Hybrid Context-Aware Fine-Grained Word Sense Disambiguation in Topic Modeling based Document Representation (2021)
  • Author(s): Wenbo Li, Einoshin Suzuki
  • Journal Title: Information Processing & Management Volume: 58 Issue: 4 Pages: 102592-102592
  • DOI: 10.1016/j.ipm.2021.102592
  • Peer Reviewed / Open Access
• [Journal Article] Detecting Anomalies from Human Activities by an Autonomous Mobile Robot Based on "Fast and Slow" Thinking (2021)
  • Author(s): Muhammad Fikko Fadjrimiratno, Yusuke Hatae, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Proc. Sixteenth International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2021), Vol. 5: VISAPP Volume: 5 Pages: 943-953
  • Peer Reviewed / Open Access
• [Journal Article] Hybrid Context-Aware Word Sense Disambiguation in Topic Modeling based Document Representation (2020)
  • Author(s): Wenbo Li, Einoshin Suzuki
  • Journal Title: Proc. 2020 IEEE International Conference on Data Mining (ICDM 2020) Volume: 1 Pages: 332-341
  • DOI: 10.1109/icdm50108.2020.00042
  • Peer Reviewed / Open Access
• [Journal Article] Detecting Outliers with One-Class Selective Transfer Machine (2020)
  • Author(s): Hirofumi Fujita, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Knowledge and Information Systems Volume: 62 Issue: 5 Pages: 1781-1818
  • DOI: 10.1007/s10115-019-01407-5
  • Peer Reviewed / Open Access
• [Journal Article] Detecting Anomalous Regions from an Image Based on Deep Captioning (2020)
  • Author(s): Yusuke Hatae, Qingpu Yang, Muhammad Fikko Fadjrimiratno, Yuanyuan Li, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Proc. VISIGRAPP 2020, Vol. 5: VISAPP Volume: 5 Pages: 326-335
  • DOI: 10.5220/0008949603260335
  • Peer Reviewed / Open Access
• [Journal Article] Context-Aware Latent Dirichlet Allocation for Topic Segmentation (2020)
  • Author(s): Wenbo Li, Tetsu Matsukawa, Hiroto Saigo, Einoshin Suzuki
  • Journal Title: Proc. 24th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2020) Volume: in press
  • Peer Reviewed / Open Access
• [Journal Article] Experimental Evaluation of GAN-Based One-Class Anomaly Detection on Office Monitoring (2020)
  • Author(s): Ning Dong, Yusuke Hatae, Muhammad Fikko Fadjrimiratno, Tetsu Matsukawa, and Einoshin Suzuki
  • Journal Title: Foundations of Intelligent Systems, LNCS (ISMIS 2020) Volume: in press
  • Peer Reviewed / Open Access
• [Journal Article] Harnessing GAN with Metric Learning for One-Shot Generation on a Fine-Grained Category (2019)
  • Author(s): Yusuke Ohtsubo, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Proc. 31st International Conference on Tools with Artificial Intelligence (ICTAI 2019) Volume: - Pages: 891-898
  • DOI: 10.1109/ictai.2019.00126
  • Peer Reviewed / Open Access
• [Journal Article] 弱教師つきデータ集合を用いるファッションスタイルの特徴学習に関する実験的評価 (2019)
  • Author(s): 本藤 拳也，松川徹，鈴木英之進
  • Journal Title: 火の国情報シンポジウム2019 Volume: -
  • Open Access
• [Journal Article] Multimodal Deep Neural Network with Image Sequence Features for Video Captioning (2018)
  • Author(s): Soichiro Oura, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Proc. 2018 International Joint Conference on Neural Networks (IJCNN 2018) Volume: - Pages: 3296-3302
  • DOI: 10.1109/ijcnn.2018.8489668
  • Peer Reviewed / Open Access
• [Journal Article] Retraining: A Simple Way to Improve the Ensemble Accuracy of Deep Neural Networks for Image Classification (2018)
  • Author(s): Kaikai Zhao, Tetsu Matsukawa, Einoshin Suzuki
  • Journal Title: Proc. 25th International Conference on Pattern Recognition (ICPR 2018) Volume: - Pages: 860-867
  • DOI: 10.1109/icpr.2018.8545535
  • Peer Reviewed / Open Access
• [Presentation] Detecting Anomalies from Human Activities by an Autonomous Mobile Robot Based on "Fast and Slow" Thinking (2021)
  • Author(s): Muhammad Fikko Fadjrimiratno
  • Organizer: Sixteenth International Conference on Computer Vision Theory and Applications (VISAPP 2021)
  • Int'l Joint Research
• [Presentation] Adversarial Minority-Class Re-Sampling for Imbalanced Sequence Classification (2021)
  • Author(s): Shin Ando
  • Organizer: Tenth International Conference on Pattern Recognition Applications and Methods (ICPRAM 2021)
  • Int'l Joint Research
• [Presentation] Visually-Private Scene Classification with Agent-collected Weak-labels (2021)
  • Author(s): Shin Ando
  • Organizer: Thirteenth International Conference on Agents and Artificial Intelligence (ICAART 2021)
  • Int'l Joint Research
• [Presentation] Hybrid Context-Aware Word Sense Disambiguation in Topic Modeling based Document Representation (2020)
  • Author(s): Wenbo Li
  • Organizer: 2020 IEEE International Conference on Data Mining (ICDM 2020)
  • Int'l Joint Research
• [Presentation] Detecting Anomalous Regions from an Image Based on Deep Captioning (2020)
  • Author(s): Yusuke Hatae, Qingpu Yang, Muhammad Fikko Fadjrimiratno, Yuanyuan Li, Tetsu Matsukawa, Einoshin Suzuki
  • Organizer: Fifteenth International Conference on Computer Vision Theory and Applications (VISAPP 2020))
  • Int'l Joint Research
• [Presentation] Harnessing GAN with Metric Learning for One-Shot Generation on a Fine-Grained Category (2019)
  • Author(s): Yusuke Ohtsubo, Tetsu Matsukawa, Einoshin Suzuki
  • Organizer: 31st International Conference on Tools with Artificial Intelligence (ICTAI 2019)
  • Int'l Joint Research
• [Presentation] 弱教師つきデータ集合を用いるファッションスタイルの特徴学習に関する実験的評価 (2019)
  • Author(s): 本藤 拳也
  • Organizer: 火の国情報シンポジウム2019
• [Presentation] Multimodal Deep Neural Network with Image Sequence Features for Video Captioning (2018)
  • Author(s): Einoshin Suzuki
  • Organizer: 2018 International Joint Conference on Neural Networks (IJCNN 2018)
  • Int'l Joint Research
• [Presentation] Retraining: A Simple Way to Improve the Ensemble Accuracy of Deep Neural Networks for Image Classification (2018)
  • Author(s): Kaikai Zhao
  • Organizer: 25th International Conference on Pattern Recognition (ICPR 2018)
  • Int'l Joint Research
• [Remarks] ストリームデータからの弱ラベル情報を用いる異常・例外パターンマイニング
  • URL: http://www.i.kyushu-u.ac.jp/~suzuki/kaken1820-j.html
• [Remarks] Anomoulous/Exceptional Pattern Mining
  • URL: http://www.i.kyushu-u.ac.jp/~suzuki/kaken1820.html