Linear Speedup in Saddle-Point Escape for Decentralized Non-Convex Optimization | IEEE Conference Publication | IEEE Xplore