RelativeMahalanobis#

class scio.scores.RelativeMahalanobis(*, act_norm=None, mode='raw', epsilon=0.0, fgm_norm=inf, weights=None)[source]#

Bases: DeepMahalanobis

RelativeMahalanobis for classification.

Functionality is implemented in DeepMahalanobis.

Parameters:

epsilon – See DeepMahalanobis.
fgm_norm – See DeepMahalanobis.
weights – See DeepMahalanobis.
mode – See BaseScoreClassif.
act_norm – See BaseScore.

References

[RFL+21]

Jie Ren, Stanislav Fort, Jeremiah Zhe Liu, Abhijit Guha Roy, Shreyas Padhy, and Balaji Lakshminarayanan. A simple fix to Mahalanobis distance for improving near-OoD detection. In ICML workshop: Uncertainty & Robustness in Deep Learning. Poster. 2021. URL: http://www.gatsby.ucl.ac.uk/~balaji/udl2021/accepted-papers/UDL2021-paper-007.pdf.