Deep Isometric Maps

作者：

Highlights：

• We present a distance preserving manifold learning scheme using neural networks

• No dense matrices and large spectral decompositions

• Landmark selection based on geodesic sampling enables faster training.

• Superior local and non-local generalization to out-of-sample points

• Tested on different challenging scenarios like noise, non-convexities like holes, conformal distortions etc.

摘要

•We present a distance preserving manifold learning scheme using neural networks•No dense matrices and large spectral decompositions•Landmark selection based on geodesic sampling enables faster training.•Superior local and non-local generalization to out-of-sample points•Tested on different challenging scenarios like noise, non-convexities like holes, conformal distortions etc.

论文关键词：Multidimensional scaling,Manifold learning,Non-linear dimensionality reduction,Neural networks

论文评审过程：Received 30 January 2021, Revised 1 March 2022, Accepted 14 April 2022, Available online 25 April 2022, Version of Record 10 May 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104461