| 摘要 |
PROBLEM TO BE SOLVED: To provide a forming method of an experimental design, which estimates a target function with the small number of experiment times, by designating an efficient experiment procedure through the user of active learning at the time of performing an experiment where all trials becoming objects cannot practically be executed. SOLUTION: In an input means, initial data, the initial distribution of data and a low-order learning algorithm are inputted. A boosting device 100 executes a processing, obtains a supposition, calculates a prediction value to present data from the supposition, calculates an error from the predicted value and updates the distribution of data. A means 210 uses the supposition which is finally obtained in the boosting device 10, selects a questing point which is not contained in past data and gives the instruction o the question point as a new experiment to an experimenter. A means 220 receives a function value obtained by the experiment of the experimenter against the question point. A means 230 adds the obtained question point and the function value to the past data. When the number of question times reaches a prescribed value in a means 240, active learning is terminated. Then, the supposition obtained at that time becomes that which is to be obtained. |
