| 摘要 |
<P>PROBLEM TO BE SOLVED: To achieve machine learning at a low cost without increasing a computer cluster. <P>SOLUTION: A second computer transmits a set of a meta parameter of learning and a feature index to a plurality of first computers, allocates a Map processing part, allocates data of a storage to perform learning processing, makes learning processing be executed as a first worker, allocates a Reduce processing part for receiving output of the Map processing part and updating a model parameter to at least one first computer, and makes the learning processing be executed as a second worker. The first worker executes feature selection and a map function to the meta parameter and the feature index of a plurality of model management parts every time a feature amount is read, calculates an intermediate result, and transmits it to the second worker. The second worker executes a Reduce function and generates a learning result from the intermediate result, and the second computer transmits the learning result to the second worker to repeat learning processing until the generated learning result satisfies a prescribed reference. <P>COPYRIGHT: (C)2013,JPO&INPIT |
