Alert button

Decentralized Riemannian Gradient Descent on the Stiefel Manifold

Feb 14, 2021
Shixiang Chen, Alfredo Garcia, Mingyi Hong, Shahin Shahrampour

Figure 1 for Decentralized Riemannian Gradient Descent on the Stiefel Manifold
Figure 2 for Decentralized Riemannian Gradient Descent on the Stiefel Manifold
Figure 3 for Decentralized Riemannian Gradient Descent on the Stiefel Manifold
Figure 4 for Decentralized Riemannian Gradient Descent on the Stiefel Manifold

Share this with someone who'll enjoy it:

We consider a distributed non-convex optimization where a network of agents aims at minimizing a global function over the Stiefel manifold. The global function is represented as a finite sum of smooth local functions, where each local function is associated with one agent and agents communicate with each other over an undirected connected graph. The problem is non-convex as local functions are possibly non-convex (but smooth) and the Steifel manifold is a non-convex set. We present a decentralized Riemannian stochastic gradient method (DRSGD) with the convergence rate of $\mathcal{O}(1/\sqrt{K})$ to a stationary point. To have exact convergence with constant stepsize, we also propose a decentralized Riemannian gradient tracking algorithm (DRGTA) with the convergence rate of $\mathcal{O}(1/K)$ to a stationary point. We use multi-step consensus to preserve the iteration in the local (consensus) region. DRGTA is the first decentralized algorithm with exact convergence for distributed optimization on Stiefel manifold.

View paper onarxiv icon

Share this with someone who'll enjoy it: