from Hacker News

Xplique Is a Neural Networks Explainability Toolbox

by AdilZtn on 2/2/24, 2:35 PM with 3 comments

  • by anto-poc on 2/2/24, 2:59 PM

    This is an explanainability library to unveil information on neural networks.

    It proposes a large variety of attribution methods, they support diverse data types (Tabular, Time Series, and Images) and diverse tasks (classification, regression, object detection, semantic segmentation).

    You should have a look to the tutorials or the documentation to know more.

  • by kyumeo on 2/2/24, 2:44 PM

    Hi!

    One of the main contributors here!

    Xplique handles both TF and Torch frameworks for a variety of attribution methods (some of them developed by our team), concept methods (including the excellent CRAFT by Thomas Fel et al.) and Feature Visualization.

    Please take a look and tell us what you think about it, any feedback welcome!

  • by raphaelty on 2/2/24, 2:50 PM

    The library is clean and well documented