| 摘要 |
PROBLEM TO BE SOLVED: To calculate a solution of a combination bandit problem at high speed and efficiently.SOLUTION: A solution search system comprises a record superiority comparison section 1, a control section 3 and an output instruction section 4. When searching a combination of examination objects 5 to which output of a best result is expected, among two or more examination objects each outputting a result on the basis of preset probability distribution, the record superiority comparison section 1 calculates a record up to the moment based on accumulation of outputted results for each examination object 5 and compares superiority in records of the examination objects 5 in a relation with records of all the examination objects 5. On the basis of the compared superiority in the records and latest results outputted from the examination objects 5, the control section 3 performs control to increase or decrease a measuring parameter for each examination object 5. The output instruction section 4 instructs an examination object 5 of which the measuring parameter exceeds a threshold value, to output a result. The output instruction section 4 specifies a combination of examination objects 5 to which the output of the result is most instructed finally after repeating the instruction of output of the result, as a combination to be searched. |
