Convex clustering is the reformulation of k-means clustering as a convex problem. While the two problems are not equivalent, the former can be seen as a relaxation of the latter that allows us to easily find globally optimal solutions (as opposed to only locally optimal ones).
Suppose we have a ...
Supervised machine learning problems are typically of the form "minimize your error while regularizing your parameters." The idea is that while many choices of parameters may make your training error low, the goal isn't low training error -- it's low test-time error. Thus, parameters should be minimize training error ...
In the mid-1980s, Yurii Nesterov hit the equivalent of an academic home run.
At the same time, he established the Accelerated Gradient Method, proved that
its convergence rate superior to Gradient Descent (
← newer continue →