• Login
    View Item 
    •   Eurographics DL Home
    • Eurographics Partner Events
    • EuroVisShort
    • EuroVisShort2021
    • View Item
    •   Eurographics DL Home
    • Eurographics Partner Events
    • EuroVisShort
    • EuroVisShort2021
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Loss-contribution-based in situ Visualization for Neural Network Training

    Thumbnail
    View/Open
    001-005.pdf (1.624Mb)
    1019-file1.mp4 (11.75Mb)
    Date
    2021
    Author
    Lee, Teng-Yok ORCID
    Pay-Per-View via TIB Hannover:

    Try if this item/paper is available.

    Metadata
    Show full item record
    Abstract
    This paper presents an in situ visualization algorithm for neural network training. As each training data item leads to multiple hidden variables when being forward-propagated through a neural network, our algorithm first estimates how much each hidden variable contributes to the training loss. Based on linear approximation, we can approximate the contribution mainly based on the forward-propagated value and the backward-propagated derivative per hidden variable, both of which are available during the training with no cost. By aggregating the loss contribution of hidden variables per data item, we can detect difficult data items that contribute most to the loss, which can be ambiguous or even incorrectly labeled. For convolution neural networks (CNN) with images as inputs, we extend the estimation of loss contribution to measure how different image areas impact the loss, which can be visualized over time to see how a CNN evolves to handle ambiguous images.
    BibTeX
    @inproceedings {10.2312:evs.20211046,
    booktitle = {EuroVis 2021 - Short Papers},
    editor = {Agus, Marco and Garth, Christoph and Kerren, Andreas},
    title = {{Loss-contribution-based in situ Visualization for Neural Network Training}},
    author = {Lee, Teng-Yok},
    year = {2021},
    publisher = {The Eurographics Association},
    ISBN = {978-3-03868-143-4},
    DOI = {10.2312/evs.20211046}
    }
    URI
    https://doi.org/10.2312/evs.20211046
    https://diglib.eg.org:443/handle/10.2312/evs20211046
    Collections
    • EuroVisShort2021

    Eurographics Association copyright © 2013 - 2023 
    Send Feedback | Contact - Imprint | Data Privacy Policy | Disable Google Analytics
    Theme by @mire NV
    System hosted at  Graz University of Technology.
    TUGFhA
     

     

    Browse

    All of Eurographics DLCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

    My Account

    LoginRegister

    Statistics

    View Usage Statistics

    BibTeX | TOC

    Create BibTeX Create Table of Contents

    Eurographics Association copyright © 2013 - 2023 
    Send Feedback | Contact - Imprint | Data Privacy Policy | Disable Google Analytics
    Theme by @mire NV
    System hosted at  Graz University of Technology.
    TUGFhA