Full text
| |
| Author(s): |
Henrique Koji Miyamoto
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Matemática, Estatística e Computação Científica |
| Defense date: | 2022-09-14 |
| Examining board members: |
Sueli Irene Rodrigues Costa;
João Eloir Strapasson;
Charles Casimiro Cavalcante
|
| Advisor: | Sueli Irene Rodrigues Costa |
| Abstract | |
This dissertation is composed of three contributions, which have in common the use of tools from geometry and/or statistics in applications to communications and learning. The first of them concerns the construction of spherical codes from a recursive procedure based on the sphere foliations given by the Hopf fibration. In the second one, we propose a method for lossy vector compression, formed by a data-adapted quantiser, followed by compression of the quantisation indices with a context-tree algorithm. The third consists in using a loss function based on the Fisher-Rao distance in the manifold of discrete distributions for training neural networks, particularly under label noise (AU) | |
| FAPESP's process: | 21/04516-8 - Information Geometry: applications to signal processing and machine learning |
| Grantee: | Henrique Koji Miyamoto |
| Support Opportunities: | Scholarships in Brazil - Master |