Coordinate descent

Coordinate descent is based on the idea that the minimization of a multivariable function ${\displaystyle F(\mathbf {x} )}$ can be achieved by minimizing it along one direction at a time, i.e., solving univariate (or at least much simpler) optimization problems in a loop.[1] In the simplest case of cyclic coordinate descent, one cyclically iterates through the directions, one at a time, minimizing the objective function with respect to each coordinate direction at a time. That is, starting with initial variable values

${\displaystyle \mathbf {x} ^{0}=(x_{1}^{0},\ldots ,x_{n}^{0})}$,

round ${\displaystyle k+1}$ defines ${\displaystyle \mathbf {x} ^{k+1}}$ from ${\displaystyle \mathbf {x} ^{k}}$ by iteratively solving the single variable optimization problems

${\displaystyle x_{i}^{k+1}={\underset {y\in \mathbb {R} }{\operatorname {arg\,min} }}\;f(x_{1}^{k+1},\dots ,x_{i-1}^{k+1},y,x_{i+1}^{k},\dots ,x_{n}^{k})}$[2]

for each variable ${\displaystyle x_{i}}$ of ${\displaystyle \mathbf {x} }$, for ${\displaystyle i}$ from 1 to ${\displaystyle n}$.

Thus, one begins with an initial guess ${\displaystyle \mathbf {x} ^{0}}$ for a local minimum of ${\displaystyle F}$, and gets a sequence ${\displaystyle \mathbf {x} ^{0},\mathbf {x} ^{1},\mathbf {x} ^{2},\dots }$ iteratively.

By doing line search in each iteration, one automatically has

${\displaystyle F(\mathbf {x} ^{0})\geq F(\mathbf {x} ^{1})\geq F(\mathbf {x} ^{2})\geq \dots .}$

It can be shown that this sequence has similar convergence properties as steepest descent. No improvement after one cycle of line search along coordinate directions implies a stationary point is reached.

This process is illustrated below.

Coordinate descent has two problems. One of them is having a non-smooth multivariable function. The following picture shows that coordinate descent iteration may get stuck at a non-stationary point if the level curves of a function are not smooth. Suppose that the algorithm is at the point (-2, -2); then there are two axis-aligned directions it can consider for taking a step, indicated by the red arrows. However, every step along these two directions will increase the objective function's value (assuming a minimization problem), so the algorithm will not take any step, even though both steps together would bring the algorithm closer to the optimum. While this example shows that coordinate descent is not necessarily convergent to the optimum, it is possible to show formal convergence under reasonable conditions.[3]

The other problem is difficulty in parallelism. Since the nature of coordinate descent is to cycle through the directions and minimize the objective function with respect to each coordinate direction, coordinate descent is not an obvious candidate for massive parallelism. Recent research works have shown that massive parallelism is applicable to coordinate descent by relaxing the change of the objective function with respect to each coordinate direction.[4][5][6]

Coordinate descent algorithms are popular with practitioners owing to their simplicity, but the same property has led optimization researchers to largely ignore them in favor of more interesting (complicated) methods.[1] An early application of coordinate descent optimization was in the area of computed tomography[7] where it has been found to have rapid convergence[8] and was subsequently used for clinical multi-slice helical scan CT reconstruction.[9] Moreover, there has been increased interest in the use of coordinate descent with the advent of large-scale problems in machine learning, where coordinate descent has been shown competitive to other methods when applied to such problems as training linear support vector machines[10] (see LIBLINEAR) and non-negative matrix factorization.[11] They are attractive for problems where computing gradients is infeasible, perhaps because the data required to do so are distributed across computer networks.[12]

• Adaptive coordinate descent
• Conjugate gradient
• Gradient descent
• Line search
• Mathematical optimization
• Newton's method
• Stochastic gradient descent – uses one example at a time, rather than one coordinate

• Bezdek, J. C.; Hathaway, R. J.; Howard, R. E.; Wilson, C. A.; Windham, M. P. (1987), "Local convergence analysis of a grouped variable version of coordinate descent", Journal of Optimization Theory and Applications, Kluwer Academic/Plenum Publishers, 54 (3), pp. 471–477, doi:10.1007/BF00940196
• Bertsekas, Dimitri P. (1999). Nonlinear Programming, Second Edition Athena Scientific, Belmont, Massachusetts. ISBN 1-886529-00-0.
• Canutescu, AA; Dunbrack, RL (2003), "Cyclic coordinate descent: A robotics algorithm for protein loop closure.", Protein Science, 12 (5), pp. 963–72, doi:10.1110/ps.0242703, PMC 2323867, PMID 12717019.
• Luo, Zhiquan; Tseng, P. (1992), "On the convergence of the coordinate descent method for convex differentiable minimization", Journal of Optimization Theory and Applications, Kluwer Academic/Plenum Publishers, 72 (1), pp. 7–35, doi:10.1007/BF00939948, hdl:1721.1/3164.
• Wu, TongTong; Lange, Kenneth (2008), "Coordinate descent algorithms for Lasso penalized regression", The Annals of Applied Statistics, Institute of Mathematical Statistics, 2 (1), pp. 224–244, arXiv:0803.3876, doi:10.1214/07-AOAS147.
• Richtarik, Peter; Takac, Martin (April 2011), "Iteration complexity of randomized block-coordinate descent methods for minimizing a composite function", Mathematical Programming, Springer, 144 (1–2), pp. 1–38, arXiv:1107.2848, doi:10.1007/s10107-012-0614-z.
• Richtarik, Peter; Takac, Martin (December 2012), "Parallel coordinate descent methods for big data optimization", ArXiv:1212.0873, arXiv:1212.0873.