Conceitos essenciais
コービットフィルターバンクは、最大フィルターバンクを一般化し、注目される。
Resumo
この論文は、実際にグループの構成要素を実現し、連続性を示すことに焦点を当てています。最初に、グループの構成要素を明確に定義し、その後、コービットマップが連続であることを証明します。さらに、最大フィルタリングの特性も再確認されます。
Introduction:
- Coorbit filter banks unify previous notions.
- Addressing ambiguity in machine learning algorithms.
Construction and Basic Properties of Coorbit Maps:
- Component coorbit map is invariant.
- Symmetry and scalar homogeneity properties are established.
Realizing Group Components and Continuity of Coorbit Maps:
- Theorem proves continuity of the coorbit map.
- Separation scale plays a crucial role in analyzing group components.
Preliminary on Max Filtering:
- Max filtering properties are recalled for further analysis.
Estatísticas
Given a real inner product space V and a group G of linear isometries, we construct a family of G-invariant real-valued functions on V that we call coorbit filter banks. When V = Rd and G is compact, a suitable coorbit filter bank is injective and locally lower Lipschitz in the quotient metric at orbits of maximal dimension. Furthermore, when the orbit space Sd−1/G is a Riemannian orbifold, a suitable coorbit filter bank is bi-Lipschitz in the quotient metric.
In this paper, we give a construction of coorbit filter banks for all compact groups G ≤O(d). These maps unify the family of max filter banks with the family of coorbit filter banks. It remains open whether every injective coorbit filter bank is bi-Lipschitz. We study whether these maps are bi-Lipschitz given enough generic templates.
Citações
"Neglecting ambiguities can magnify sample complexity."
"Every injective coorbit filter bank admits bi-Lipschitz bounds."
"Co-orbit maps enjoy semialgebraic property."