zoNNscan: a boundary-entropy index for zone inspection of neural models - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year :

zoNNscan: a boundary-entropy index for zone inspection of neural models

Abstract

The training of deep neural network classifiers results in decision boundaries whose geometry is still not well understood. This is in direct relation with classification problems such as so called corner case inputs. We introduce zoNNscan, an index that is intended to inform on the boundary uncertainty (in terms of the presence of other classes) around one given input datapoint. It is based on confidence entropy, and is implemented through Monte Carlo sampling in the multidimensional ball surrounding that input. We detail the zoNNscan index, give an algorithm for approximating it, and finally illustrate its benefits on three applications.
Fichier principal
Vignette du fichier
main.pdf (501.82 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03118264 , version 1 (22-01-2021)

Identifiers

  • HAL Id : hal-03118264 , version 1

Cite

Adel Jaouen, Erwan Le Merrer. zoNNscan: a boundary-entropy index for zone inspection of neural models. MCS 2020 - Monte Carlo Search workshop, Jan 2021, Virtual, Japan. pp.1-8. ⟨hal-03118264⟩
36 View
43 Download

Share

Gmail Facebook Twitter LinkedIn More