An accelerated version of Multiplicative Weights Update that avoids saddle points

Home

Discover

Bookmarks

Keep Up With Top Trending Papers In CS / AI / ML ...

Optimization and Control

An accelerated version of Multiplicative Weights Update that avoids saddle points

Accelerated Multiplicative Weights Update Avoids Saddle Points almost always

We provide an accelerated version of the multiplicative weights update (mwu) based on the riemannian accelerated gradient descent, and prove that the riemannian accelerated gradient descent, thus theaccelerated version of the multiplicative weights update (mwu), almost always avoid saddle points.

Multiplicative weights update, riemannian accelerated gradient descent, nonconvex optimization.

Authors

Yi Feng, Ioannis Panageas, Xiao Wang

◐ Recommended Members

◐ Latest Activity