Adaptive Drawing Behavior by Visuomotor Learning Using Recurrent Neural Networks

Kazuma Sasaki, Tetsuya Ogata

    Research output: Contribution to journalArticle

    Abstract

    Drawing is a medium that represents an idea as drawn lines, and drawing behavior requires complex cognitive abilities to process visual and motor information. One way to understand aspects of these abilities is constructing computational models that can replicate these abilities rather than explaining the phenomena by building plausible models by a top-down manner. In this study, we proposed a supervised learning model that can be trained using examples of visuomotor sequences from drawings made by human. Additionally, we demonstrated that the proposed model has functions of 1) associating motions to depict the given picture image and 2) adapting to drawing behavior to complete a given part of the drawing process. This dynamical model is implemented by recurrent neural networks that have images and motion as their input and output. Through experiments that involved learning human drawing sequences, the model was able to associate appropriate motions to achieve depiction targets while adapting to a given part of the drawing process. Furthermore, we demonstrate that including visual information in the model improved performance robustness against noisy lines in the input data.

    Original languageEnglish
    JournalIEEE Transactions on Cognitive and Developmental Systems
    DOIs
    Publication statusAccepted/In press - 2018 Aug 31

    Fingerprint

    Recurrent neural networks
    Supervised learning

    Keywords

    • Adaptation
    • Adaptation models
    • Computational modeling
    • Drawing ability
    • Predictive models
    • Recurrent neural networks
    • Recurrent Neural Networks
    • Supervised learning
    • Training
    • Visualization
    • Visuomotor Learning.

    ASJC Scopus subject areas

    • Software
    • Artificial Intelligence

    Cite this

    @article{a83f3b91d4774c92840b380b0ac60186,
    title = "Adaptive Drawing Behavior by Visuomotor Learning Using Recurrent Neural Networks",
    abstract = "Drawing is a medium that represents an idea as drawn lines, and drawing behavior requires complex cognitive abilities to process visual and motor information. One way to understand aspects of these abilities is constructing computational models that can replicate these abilities rather than explaining the phenomena by building plausible models by a top-down manner. In this study, we proposed a supervised learning model that can be trained using examples of visuomotor sequences from drawings made by human. Additionally, we demonstrated that the proposed model has functions of 1) associating motions to depict the given picture image and 2) adapting to drawing behavior to complete a given part of the drawing process. This dynamical model is implemented by recurrent neural networks that have images and motion as their input and output. Through experiments that involved learning human drawing sequences, the model was able to associate appropriate motions to achieve depiction targets while adapting to a given part of the drawing process. Furthermore, we demonstrate that including visual information in the model improved performance robustness against noisy lines in the input data.",
    keywords = "Adaptation, Adaptation models, Computational modeling, Drawing ability, Predictive models, Recurrent neural networks, Recurrent Neural Networks, Supervised learning, Training, Visualization, Visuomotor Learning.",
    author = "Kazuma Sasaki and Tetsuya Ogata",
    year = "2018",
    month = "8",
    day = "31",
    doi = "10.1109/TCDS.2018.2868160",
    language = "English",
    journal = "IEEE Transactions on Cognitive and Developmental Systems",
    issn = "2379-8920",
    publisher = "Institute of Electrical and Electronics Engineers Inc.",

    }

    TY - JOUR

    T1 - Adaptive Drawing Behavior by Visuomotor Learning Using Recurrent Neural Networks

    AU - Sasaki, Kazuma

    AU - Ogata, Tetsuya

    PY - 2018/8/31

    Y1 - 2018/8/31

    N2 - Drawing is a medium that represents an idea as drawn lines, and drawing behavior requires complex cognitive abilities to process visual and motor information. One way to understand aspects of these abilities is constructing computational models that can replicate these abilities rather than explaining the phenomena by building plausible models by a top-down manner. In this study, we proposed a supervised learning model that can be trained using examples of visuomotor sequences from drawings made by human. Additionally, we demonstrated that the proposed model has functions of 1) associating motions to depict the given picture image and 2) adapting to drawing behavior to complete a given part of the drawing process. This dynamical model is implemented by recurrent neural networks that have images and motion as their input and output. Through experiments that involved learning human drawing sequences, the model was able to associate appropriate motions to achieve depiction targets while adapting to a given part of the drawing process. Furthermore, we demonstrate that including visual information in the model improved performance robustness against noisy lines in the input data.

    AB - Drawing is a medium that represents an idea as drawn lines, and drawing behavior requires complex cognitive abilities to process visual and motor information. One way to understand aspects of these abilities is constructing computational models that can replicate these abilities rather than explaining the phenomena by building plausible models by a top-down manner. In this study, we proposed a supervised learning model that can be trained using examples of visuomotor sequences from drawings made by human. Additionally, we demonstrated that the proposed model has functions of 1) associating motions to depict the given picture image and 2) adapting to drawing behavior to complete a given part of the drawing process. This dynamical model is implemented by recurrent neural networks that have images and motion as their input and output. Through experiments that involved learning human drawing sequences, the model was able to associate appropriate motions to achieve depiction targets while adapting to a given part of the drawing process. Furthermore, we demonstrate that including visual information in the model improved performance robustness against noisy lines in the input data.

    KW - Adaptation

    KW - Adaptation models

    KW - Computational modeling

    KW - Drawing ability

    KW - Predictive models

    KW - Recurrent neural networks

    KW - Recurrent Neural Networks

    KW - Supervised learning

    KW - Training

    KW - Visualization

    KW - Visuomotor Learning.

    UR - http://www.scopus.com/inward/record.url?scp=85052840258&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85052840258&partnerID=8YFLogxK

    U2 - 10.1109/TCDS.2018.2868160

    DO - 10.1109/TCDS.2018.2868160

    M3 - Article

    JO - IEEE Transactions on Cognitive and Developmental Systems

    JF - IEEE Transactions on Cognitive and Developmental Systems

    SN - 2379-8920

    ER -