Random coordinate descent

Randomized (Block) Coordinate Descent Method is an optimization algorithm popularized by Nesterov (2010) and Richtárik and Takáč (2011). The first analysis of this method, when applied to the problem of minimizing a smooth convex function, was performed by Nesterov (2010).[1] In Nesterov's analysis the method needs to be applied to a quadratic perturbation of the original function with an unknown scaling factor. Richtárik and Takáč (2011) give iteration complexity bounds which do not require this, i.e., the method is applied to the objective function directly. Furthermore, they generalize the setting to the problem of minimizing a composite function, i.e., sum of a smooth convex and a (possibly nonsmooth) convex block-separable function:

where is decomposed into blocks of variables/coordinates: and are (simple) convex functions.

Example (block decomposition): If and , one may choose and .

Example (block-separable regularizers):

  1. , where and is the standard Euclidean norm.

Algorithm

edit

Consider the optimization problem

 

where   is a convex and smooth function.

Smoothness: By smoothness we mean the following: we assume the gradient of   is coordinate-wise Lipschitz continuous with constants  . That is, we assume that

 

for all   and  , where   denotes the partial derivative with respect to variable  .

Nesterov, and Richtarik and Takac showed that the following algorithm converges to the optimal point:

Algorithm Random Coordinate Descent Method
    Input:   //starting point
    Output:  

    set x := x_0

    for k := 1, ... do
        choose coordinate  , uniformly at random
        update   
    end for
  • "←" denotes assignment. For instance, "largestitem" means that the value of largest changes to the value of item.
  • "return" terminates the algorithm and outputs the following value.

Convergence rate

edit

Since the iterates of this algorithm are random vectors, a complexity result would give a bound on the number of iterations needed for the method to output an approximate solution with high probability. It was shown in [2] that if  , where  ,   is an optimal solution ( ),   is a confidence level and   is target accuracy, then  .

Example on particular function

edit

The following Figure shows how   develops during iterations, in principle. The problem is

 

 

Extension to block coordinate setting

edit
 
Blocking coordinate directions into Block coordinate directions

One can naturally extend this algorithm not only just to coordinates, but to blocks of coordinates. Assume that we have space  . This space has 5 coordinate directions, concretely   in which Random Coordinate Descent Method can move. However, one can group some coordinate directions into blocks and we can have instead of those 5 coordinate directions 3 block coordinate directions (see image).

See also

edit

References

edit
  1. ^ Nesterov, Yurii (2010), "Efficiency of coordinate descent methods on huge-scale optimization problems", SIAM Journal on Optimization, 22 (2): 341–362, CiteSeerX 10.1.1.332.3336, doi:10.1137/100802001
  2. ^ Richtárik, Peter; Takáč, Martin (2011), "Iteration complexity of randomized block-coordinate descent methods for minimizing a composite function", Mathematical Programming, Series A, 144 (1–2): 1–38, arXiv:1107.2848, doi:10.1007/s10107-012-0614-z