Personality and meta-learning in terms of neural transferable factors

• Project/Area Number: 19H04180
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Kyoto University
• Principal Investigator: Ishii Shin 京都大学, 情報学研究科, 教授 (90294280)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Project Status: Completed (Fiscal Year 2022)
• Budget Amount: ¥17,290,000 (Direct Cost: ¥13,300,000、Indirect Cost: ¥3,990,000); Fiscal Year 2021: ¥4,940,000 (Direct Cost: ¥3,800,000、Indirect Cost: ¥1,140,000); Fiscal Year 2020: ¥5,200,000 (Direct Cost: ¥4,000,000、Indirect Cost: ¥1,200,000); Fiscal Year 2019: ¥7,150,000 (Direct Cost: ¥5,500,000、Indirect Cost: ¥1,650,000)
• Keywords: 脳型人工知能 / ブレイン・マシン・インターフェース / 模倣学習 / 強化学習 / ベイズ推定 / 意思決定 / 順逆強化学習 / 敵対事例学習 / ソフトコンピューティング

Outline of Research at the Start

脳は、個体間で共通するハードウェアとしての機能単位が階層的に存在し、その使い方の違いが個性に対応するという第一の仮説の下、機能単位の転移法を導出し、被験者転移型ブレイン・マシン・インターフェースおよび個人認証技術につなげる。また、複数の機能単位の一部が環境に適応し残りの部分は再利用可能という効果的な学習能力が、ダイナミックな環境への素早い適応と安定性を同時に実現するという第二の仮説の下、マルチタスク環境の自律的学習を可能とするアルゴリズムを導出する。二つの機能単位の関連を機能的脳イメージングにより調べると共に、効果的な学習法としての工学的評価を行う。

Outline of Final Research Achievements

We clarified fMRI-based neural bases involved in bottom-up and top-down attention during a still image observation task; this result has been published in (Fujimoto, et al., 2023). Extending this approach to examine prior factors of image observers, we successfully constructed a Bayesian image recognition model that could reproduce observers’ behaviors well. With a hierarchical decision-making task, we developed a hierarchical Bayesian inference model and also examined neural bases in hierarchical uncertainty resolution within decision making.
Moreover, we developed a new autonomous learning method that combines adversarial learning and reinforcement learning for central pattern generator, and applied it to autonomous locomotion acquisition tasks by a quadruped robot similar. We found the simulator could produce stable locomotion on a variety of terrain by this method.

Academic Significance and Societal Importance of the Research Achievements

覚醒脳におけるトップダウン注意の関与を実験的に調べた研究は少なく、本研究の成果は学術的意義がある。以下のように新聞報道（オンライン版）された。
https://www.nikkei.com/article/DGXZQOUC132QF0T11C22A0000000/
また、敵対事例学習に中枢パターン生成器強化学習を組み合わせた新しい学習法は、ロボットのみならず、自動車などの人工物の制御にも利用可能と期待される。この成果は制御系で最大級の国際会議IFAC 2023の招待セッションにて発表される。

Research Products

• [Int'l Joint Research] Oxford University(英国)
• [Int'l Joint Research] 深セン大学(中国)
• [Journal Article] Deep learning-based image deconstruction method with maintained saliency (2022)
  • Author(s): Fujimoto, K., Hayashi, K., Katayama, R., Lee, S., Liang, Z., Yoshida, W., Ishii, S.
  • Journal Title: Neural Networks Volume: 155 Pages: 224-241
  • DOI: 10.1016/j.neunet.2022.08.015
  • Peer Reviewed / Open Access / Int'l Joint Research
• [Journal Article] Confidence modulates the decodability of scene prediction during partially-observable maze exploration in humans (2022)
  • Author(s): Katayama, R., Yoshida, Y., Ishii, S.
  • Journal Title: Communications Biology Volume: 5 Issue: 1 Pages: 367-367
  • DOI: 10.1038/s42003-022-03314-y
  • Peer Reviewed / Open Access / Int'l Joint Research
• [Journal Article] Tri-view two-photon microscopic image registration and deblurring with convolutional neural networks (2022)
  • Author(s): Lee Sehyung、Kume Hideaki、Urakubo Hidetoshi、Kasai Haruo、Ishii Shin
  • Journal Title: Neural Networks Volume: 152 Pages: 57-69
  • DOI: 10.1016/j.neunet.2022.04.011
  • Peer Reviewed / Open Access
• [Journal Article] Modularized Predictive Coding-Based Online Motion Synthesis Combining Environmental Constraints and Motion-Capture Data (2020)
  • Author(s): Hwang Jaepyung、Ishii Shin、Kwon Taesoo、Oba Shigeyuki
  • Journal Title: IEEE Access Volume: 8 Pages: 202274-202285
  • DOI: 10.1109/access.2020.3036449
  • Peer Reviewed
• [Journal Article] A Generalized Encoding System for Alpha Oscillations Through Visual Saliency Analysis (2020)
  • Author(s): Liang Zhen、Li Fangchao、Hu Wanrou、Huang Gan、Oba Shigeyuki、Zhang Zhiguo、Ishii Shin
  • Journal Title: IEEE Transactions on Neural Systems and Rehabilitation Engineering Volume: 28 Issue: 12 Pages: 2731-2743
  • DOI: 10.1109/tnsre.2020.3038789
  • Peer Reviewed
• [Journal Article] A Meta-Q-Learning Approach to Discriminative Correlation Filter based Visual Tracking (2020)
  • Author(s): Kubo Akihiro、Meshgi Kourosh、Ishii Shin
  • Journal Title: Journal of Intelligent & Robotic Systems Volume: 101 Issue: 1 Pages: 1-11
  • DOI: 10.1007/s10846-020-01273-2
• [Journal Article] EEG-based personal identification method using unsupervised feature extraction and its robustness against intra-subject variability (2020)
  • Author(s): Nishimoto Takashi、Higashi Hiroshi、Morioka Hiroshi、Ishii Shin
  • Journal Title: Journal of Neural Engineering Volume: 17 Issue: 2 Pages: 026007-026007
  • DOI: 10.1088/1741-2552/ab6d89
  • Peer Reviewed / Open Access
• [Journal Article] Computational characteristics of the striatal dopamine system described by reinforcement learning with fast generalization (2019)
  • Author(s): Fujita Yoshihisa、Yagishita Sho、Kasai Haruo、Ishii Shin
  • Journal Title: Frontiers in Computational Neuroscience
  • DOI: 10.1101/2019.12.12.873950
• [Journal Article] Constrained Deep Q-Learning Gradually Approaching Ordinary Q-Learning (2019)
  • Author(s): Shota Ohnishi, Eiji Uchibe, Yotaro Yamaguchi, Kosuke Nakanishi, Yuji Yasui, and Shin Ishii
  • Journal Title: Frontiers in Neurorobotics Volume: 13
  • DOI: 10.3389/fnbot.2019.00103
  • Peer Reviewed / Open Access
• [Journal Article] An unsupervised EEG decoding system for human emotion recognition (2019)
  • Author(s): Liang Zhen、Oba Shigeyuki、Ishii Shin
  • Journal Title: Neural Networks Volume: 116 Pages: 257-268
  • DOI: 10.1016/j.neunet.2019.04.003