2023 Fiscal Year Final Research Report
Applications of the study of vision and visual illusions by harmonic analysis methods to machine learning
Project/Area Number |
18K18716
|
Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
Allocation Type | Multi-year Fund |
Review Section |
Medium-sized Section 12:Analysis, applied mathematics, and related fields
|
Research Institution | Waseda University |
Principal Investigator |
Arai Hitoshi 早稲田大学, 教育・総合科学学術院, 教授 (10175953)
|
Project Period (FY) |
2018-06-29 – 2024-03-31
|
Keywords | 単純かざぐるまフレームレット / かざぐるまフレームレット / 錯視 / 深層学習 / 調和解析的方法 |
Outline of Final Research Achievements |
In the exploratory research project by this Grant-in-Aid for Scientific Research, the principal investigator, Hitoshi Arai, constructed some deep neural networks using harmonic analysis methods such as simple pinwheel framelets and pinwheel framelets established by H.Arai and S.Arai in 2009, 2011. In this exploratory research project, H. Arai also created newly a dataset of 30000 string tilt illusions of 5 characters and 30000 non-illusory character strings consisting of 5 characters. Using 70% of the dataset as training data, 10% as validation data, and 20% as test data, H. Arai classified the test data to tilt illusions and non-tilt illusions using the deep neural networks described above, and obtained a high accuracy rate of over 98%. H. Arai obtained also some new themes related to the study of visual illusions and deep learning by harmonic analysis methods that will develop this exploratory research.
|
Free Research Field |
数学,解析学
|
Academic Significance and Societal Importance of the Research Achievements |
本研究成果は調和解析,錯視科学,人工知能を結ぶ学際的研究としてその学術的意義は高い.特に文字列傾斜錯視について,研究代表者らによる数理モデル(新井仁之・新井しのぶ,特許第5456931号,2013年特許査定)とディープニューラルネットとの比較検討をすることにより,AIによる分類のメカニズムの研究につながる可能性がある.本研究で作成した文字列傾斜錯視データセットも初めてのものであり,錯視科学上有用なものである.また,浮遊錯視への可能性も得られたが,それはオプアートや商用化への可能性という社会的意義を含む.本研究成果は萌芽的なものであるが,AIによる錯視の研究に新たなテーマを切り開いたといえよう.
|