Reproduce “Non-Euclidean Broximal Point Method: A Blueprint for Geometry-Aware Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kaja Gruntkowska, Peter Richtárik
Researchers from KAUST extended the Broximal Point Method to arbitrary non-Euclidean geometries, establishing a theoretical blueprint for geometry-aware optimization. This generalization preserves several strong convergence properties, including super-accelerated linear convergence in function values, while clarifying where guarantees like monotonic distance decrease require specific norm structures.
Key insights
- The chosen non-Euclidean norm itself acts as a form of "geometric preconditioning," shaping the search space to potentially accelerate convergence.
- Many strong convergence properties of Euclidean BPM, such as one-step convergence and super-accelerated linear convergence in function values, extend to arbitrary normed spaces.
- Monotonic decrease in distance to the minimizer, a hallmark of Euclidean BPM, does not hold universally for non-Euclidean norms but can be recovered for norms induced by inner products.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.