Kn5convexe Best -

by P. Bartlett, S. Bubeck, et al. (2017) Shows that training deep networks with certain activations can be viewed as convex optimization over probability measures.

Performance degrades exponentially. Calculations that take milliseconds in 2D can take hours in 10D. kn5convexe best

In the labyrinthine world of computational geometry and data processing, the "convex hull"—the smallest convex shape that encloses a set of points—has long been a solved problem. For decades, algorithms like Quickhull and Graham Scan have been the standard bearers. But as datasets grow from gigabytes to petabytes and dimensions multiply, the old guard is creaking under the pressure. by P. Bartlett

Conversely, the implementation ensures: