Abstract
dc:descriptionSampling from distributions supported on manifolds arises in statistical models with equality constraints. Existing manifold samplers can be effective, but their performance is often sensitive to the choice of step size. This thesis proposes a novel manifold sampler, Manifold AutoStep, featuring an automatic step-size selection. The fixed-step sampler uses random-walk proposals in the tangent space followed by a projection onto the manifold. AutoStep adapts the step size of the tangent proposal locally, reducing the need for manual tuning. Theoretically, we first formulate the fixed-step manifold sampler within the framework of involutive MCMC, then apply this framework to establish the correctness of our method. In the experiments, our method is validated with a novel correctness test and is benchmarked against the fixed-step manifold sampler on several distributions with different geometric and statistical properties. Experiment results show that our method reduces sensitivity to step-size tuning and can improve exploration when suitable proposal scales vary across the state space. Overall, our method provides a practical adaptive step-size strategy for random-walk sampling from distributions on implicitly defined manifolds.
Degree
thesis:*- Name thesis:degree_name
- Master of Science - MSc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Statistics
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Tang, Junsong
- Contributors dc:contributor
-
- Bouchard-Côté, Alexandre
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95756