SLISEMAP: supervised dimensionality reduction through local explanations

Abstract Existing methods for explaining black box learning models often focus on building local explanations of the models’ behaviour particular data items. It is possible to create global all items, but these generally have low fidelity complex models. We propose a new supervised manifold visualisation method, slisemap , that simultaneously finds items and builds (typically) two-dimensional model such with similar are projected nearby. provide mathematical derivation our problem an open source implementation implemented using GPU-optimised PyTorch library. compare multiple popular dimensionality reduction find able utilise labelled embeddings consistent white also other model-agnostic explanation show provides comparable visualisations can give broader understanding regression classification