Lattice-structured secondary elastic nets for large-scaled cortex-like mapping
碩士 === 國立東華大學 === 應用數學系 === 105 === This work explores an unsupervised concurrent learning process toward organizing many second-level or secondary elastic nets for large-scaled cortex-like mapping. Fitting high-dimensional patterns to lattice-structured Gaussian mixtures naturally induces Voronoi p...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Published: |
2017
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/2523hv |
| Summary: | 碩士 === 國立東華大學 === 應用數學系 === 105 === This work explores an unsupervised concurrent learning process toward organizing many second-level or secondary elastic nets for large-scaled cortex-like mapping. Fitting high-dimensional patterns to lattice-structured Gaussian mixtures naturally induces Voronoi partition to the original feature space and neighboring relations among partitioned subspaces. Many secondary elastic nets are proposed for big data analysis and are structured and interconnected following neighboring relations of partitioned subspaces. Every partitioned subspace is equipped with one individual elastic net. Adapting nodes on one individual elastic net are constrained by boundary conditions proposed by positions of connected nodes of neighboring elastic nets as well as conventional minimal wiring and maximal fitting criteria. The proposed unsupervised learning process optimizes all secondary elastic nets simultaneously under physical-like annealing for emulating formation of essential large-scale cortex-like mapping.
|
|---|