Manifold network modeling interpretability method based on geodesic measurement
The invention discloses a manifold network modeling interpretability method based on geodesic measurement, and the method comprises the steps: carrying out the manifold feature modeling of an original image through a Riemannian manifold, and enabling an extracted feature subspace to serve as the inp...
Saved in:
Main Authors | , , , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
08.07.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention discloses a manifold network modeling interpretability method based on geodesic measurement, and the method comprises the steps: carrying out the manifold feature modeling of an original image through a Riemannian manifold, and enabling an extracted feature subspace to serve as the input of a deep learning network; and then, based on the geodesic distance of the feature subspace and a back propagation model, performing gradient model derivation on the deep learning network on the flow shape, defining a flow shape bending degree index, and finally outputting a manifold bending degree index of each layer of feature space by calculating the geodesic distance between feature layers of the deep network. According to the method, the geometric structure of data is effectively utilized, manifold modeling is carried out on the feature space of the deep features, the effectiveness principle of deep learning is analyzed from the geometric angle of the manifold space, and the interpretability method of the |
---|---|
Bibliography: | Application Number: CN202011502227 |