Optimization on Manifolds
summary
Overview of practical Riemannian optimization: definitions of manifolds (Stiefel, Grassmann, fixed-rank, SPD, hyperbolic), explicit formulas for tangent spaces, Riemannian metric, gradients and Hessians for embedded submanifolds, and computational tools that reduce to matrix factorizations. Introduces Manopt toolboxes, uses retraction maps (project or normalize or truncated SVD) to move on manifold, derives Taylor expansion of f∘retraction, and gives complexity for Riemannian gradient descent matching Euclidean rates O(1/√K).
AI-written summary · about
chapters (16)
- 0:00Start
- 4:59Romanian Manifolds
- 5:23What Is a Manifold
- 8:23The Stifle Angle
- 10:35Grass Man Manifold
- 16:32What Is the Manifold
- 20:22Why Do We Care about Manifolds
- 20:54Tangent Vector
- 32:34Metric Projection
- 33:53The Tangent Bundle
- 39:02A Vector Field on a Manifold
- 47:37Hessians
- 50:01Affine Connection
- 52:23An Algorithm on a Manifold
- 52:39Example of an Algorithm
- 55:07Proving Global Convergence Rates
related talks
11mGeometric Methods for Machine Learning and Optimization
01h 18mOptimization I
31mRandomized Riemannian Preconditioning for Quadratically Constrained Problems
51mFirst-Order Methods and Hyperbolic Programming
01h 01mRobustness, Stochastics, Uncertainty 2
49mMetric Representations: Algorithms and Geometry
46mSmooth Nonconvex Min-Max Optimization
50mUsing Dependent Types in Mathematical Optimization Algorithms
