SPARSE VARIABLE OPTIMIZATION DEVICE, SPARSE VARIABLE OPTIMIZATION METHOD, AND SPARSE VARIABLE OPTIMIZATION PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a sparse variable optimization device capable of optimizing a variable of a sparse structure by suppressing a calculation amount even when a closed form solution is not available.SOLUTION: A gradient calculation part 81 calculates a gradient of an object function in a variable as an object to be optimized. An additional variable selection part 82 adds the variable which is the greatest in absolute value of the calculated gradient among the variables included in a variable collection to a non-zero variable collection. A variable optimization part 83 optimizes a value of the variable as an object to be optimized for the variables included in the non-zero variable collection. A deletion variable selection part 84 eliminates the variable in which an increase of the object function is the smallest when the variable is eliminated from among the variables included in the non-zero variable collection, from the non-zero variable collection. An object function evaluation part 85 calculates a value of the object function for the variable as the object to be optimized.