Max-Plus Operators Applied to Filter Selection and Model Pruning in Neural Networks

Abstract :

In the context of mathematical morphology based on struc- turing elements to define erosion and dilation, this paper generalizes the notion of a structuring element to a new setting called structuring neighborhood systems. While a structuring element is often defined as a subset of the space, a structuring neighborhood is a subset of the subsets of the space. This yields an extended definition of erosion; dilation can be obtained as well by a duality principle. With respect to the classical framework, this extension is sound in many ways. It is also strictly more expressive, for any structuring element can be represented as a struc- turing neighborhood but the converse is not true. A direct application of this framework is to generalize modal morpho-logic to a topological setting.

Complete list of metadatas

https://hal.telecom-paristech.fr/hal-02288568
Contributor : Telecomparis Hal <>
Submitted on : Saturday, September 14, 2019 - 6:57:08 PM
Last modification on : Monday, September 16, 2019 - 1:09:50 AM

Identifiers

  • HAL Id : hal-02288568, version 1

Citation

Alexandre Goy, Marc Aiguier, Isabelle Bloch. Max-Plus Operators Applied to Filter Selection and Model Pruning in Neural Networks. 14th International Symposium on Mathematical Morphology (ISMM), 2019, Saarbr\"ucken, Germany. pp.16-28. ⟨hal-02288568⟩

Share

Metrics

Record views

42