the open
    research
    Contribute
    Back to the reading lists

    Safety, alignment & interpretability· 2017

    Learning Important Features Through Propagating Activation Differences

    Avanti Shrikumar (first author) · 2017

    Read the original

    arXiv 1704.02685OpenAlex W2605409611

    On the reading lists

    Why it matters

    Introduced reference-based activation attribution, addressing limitations of ordinary gradient explanations.

    Condensed from the expert lists' notes.

    The record

    First released
    2017 (from arXiv 1704.02685)
    Citations
    4,873 citations, as counted by Papers With Code (CC BY-SA 4.0)
    Named by
    both expert lists
    Score
    0.491 · computed rank 238 · consensus 0.72, must-read 0, citations percentile 0.787 (2010s · safety), awards 0, existing lists 0. Method

    Data sources

    Catalogue data from OpenAlex (CC0), Crossref (CC0 metadata; no abstracts are shown), arXiv, and Papers With Code (CC BY-SA 4.0). Each count names its source. See the method.