| 摘要 |
PROBLEM TO BE SOLVED: To provide a training system which enables a user to make effective learning with selective sampling. SOLUTION: An input device 1 inputs learning data and an end condition to which labels are set. A learning portion 21 generates hypothesis by using a learning algorithm. A storage portion 33 contains a plurality of candidate data to which labels are not set. A data selection portion 23 calculates a score for each of the plurality of candidate data by referring to the storage portion and using the hypothesis, and selects a first candidate data from the plurality of candidate data on the basis of the calculated score. A data update portion 24 sets a label determined by the user to the first candidate data selected above, and outputs the selected first candidate data together with the learning data to the learning portion. The data selection processing, the data update processing, and the learning processing are repeated until the end condition is satisfied. When the end condition is satisfied, a control portion 22 outputs the hypothesis generated by the learning portion to an output device 4, as a final hypothesis. COPYRIGHT: (C)2010,JPO&INPIT
|
