- Title
- Topological analysis, non-linear dimensionality reduction and optimisation applied to manifolds represented by point clouds
- Creator
- Paul, Rahul
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2018
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- In recent years, there has been a growing demand for computational techniques that respect the non-linear structure of high-dimensional data, in both real-world applications and research. Various forms of manifolds can describe non-linear objects. However, manifolds are abstract mathematical concepts and in applications these are often represented by high-dimensional finite sets of sample points. This thesis investigates techniques from machine learning, optimisation and computational topology that can be applied to such point clouds. The first part of this thesis presents a topological approach for validating nonlinear dimensionality reduction. During the process of non-linear dimensionality reduction, manifolds represented by point clouds are at risk of changing their topology. The impact of manifold learning is evaluated by comparing Betti numbers based on persistent homology of test manifolds before and after dimensionality reduction. The second part of the thesis addresses the processing of large point cloud data as it can occur in real applications. The topological analysis of this data using traditional methods for persistent homology can be a computationally costly task. If the data is represented by large point clouds, many current computing systems find processing difficult or fail to process the data. This thesis proposes an alternative approach that employs deep learning to estimate Betti numbers of manifolds represented by point clouds. The third part of the thesis investigates simulated examples of optimisation on general differentiable manifolds without the requirement of a Riemannian structure. A barrier method with exact line search for the optimisation problem over manifolds is proposed. The last part of this thesis reports on collaborative field work with Xerox India using a real-world data set. A heuristic algorithm is employed to solve a practical task allocation problem.
- Subject
- manifold learning; point cloud; deep learning; optimisation
- Identifier
- http://hdl.handle.net/1959.13/1393470
- Identifier
- uon:33548
- Rights
- Copyright 2018 Rahul Paul
- Language
- eng
- Full Text
- Hits: 472
- Visitors: 725
- Downloads: 243
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | ATTACHMENT01 | Thesis | 7 MB | Adobe Acrobat PDF | View Details Download | ||
| View Details Download | ATTACHMENT02 | Abstract | 210 KB | Adobe Acrobat PDF | View Details Download |