Riemannian Projection-free Online Learning

Authors: Zihao Hu, Guanghui Wang, Jacob D. Abernethy

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper, we present methods for obtaining sub-linear regret guarantees in online geodesically convex optimization on curved spaces for two scenarios: when we have access to (a) a separation oracle or (b) a linear optimization oracle. For geodesically convex losses, and when a separation oracle is available, our algorithms achieve O(T 1/2), O(T 3/4) and O(T 1/2) adaptive regret guarantees in the full information setting, the bandit setting with one-point feedback and the bandit setting with two-point feedback, respectively. When a linear optimization oracle is available, we obtain regret rates of O(T 3/4) for geodesically convex losses and O(T 2/3 log T) for strongly geodesically convex losses.
Researcher Affiliation Collaboration Zihao Hu , Guanghui Wang , Jacob Abernethy , College of Computing, Georgia Institute of Technology Google Research
Pseudocode Yes Algorithm 1: Infeasible Riemannian OGD, Algorithm 2: Infeasible Projection onto (1 δ)K with a Riemannian Separation Oracle, Algorithm 3: Infeasible R-OGD with a separation oracle, Algorithm 4: One-point bandit convex optimization on manifolds with a separation oracle, Algorithm 5: Two-point bandit convex optimization on manifolds with a separation oracle, Algorithm 6: Separating Hyperplane via RFW, Algorithm 7: Closer Infeasible Projection via LOO, Algorithm 8: Block OGD on manifolds with a linear optimization oracle, Algorithm 9: Riemannian Frank-Wolfe with line-search
Open Source Code No The paper does not provide any links to open-source code or state that the code will be released.
Open Datasets No The paper is theoretical and does not describe any experimental training on datasets.
Dataset Splits No The paper is theoretical and does not describe any experimental validation on datasets.
Hardware Specification No The paper is theoretical and does not mention any hardware specifications used for experiments.
Software Dependencies No The paper is theoretical and does not list specific software dependencies with version numbers for experimental setup.
Experiment Setup No The paper is theoretical and does not provide an experimental setup section with hyperparameters or training configurations.