US 12,112,275 B2 · Learning device, learning method and learning program

Abstract

There is provided a learning device for learning a neural network used for search of external knowledge in order to increase search accuracy of external knowledge required for arithmetic processing. With an input sentence Q as an input, an external knowledge search unit 22 selects pieces of external knowledge based on similarity degrees between pieces of external knowledge included in an external knowledge database 2 and the input sentence Q, using a neural network, and causes the selected pieces of external knowledge to be search results R2. A processing unit 14 acquires a response sentence A to the input sentence Q by arithmetic processing with the input sentence Q and the selected pieces of external knowledge as an input. A consideration calculation unit 23 calculates a consideration v determined from an index indicating correctness of the response sentence A based on a true output T given to the input sentence Q in advance and an index indicating quality of the selected pieces of external knowledge. A learning unit 26 updates neural network parameters of the second external knowledge search unit 22 using the consideration v.

Claims20

Dependent claims

A learning device comprising a processor configured to execute operations comprising:

determining scores based on similarity degrees between an input sentence and pieces of external knowledge stored in an external knowledge database;
selecting at least a part of the pieces of external knowledge based on the scores as search results;
acquiring an output to the input sentence based on predetermined arithmetic processing using the input sentence and the selected at least a part of the pieces of external knowledge as inputs;
determining, based on a combination of the input sentence, the acquired output, the selected pieces of external knowledge, and a true output given to the input sentence in advance, a reward, wherein the reward is based on a first index and a second index, the first index indicates correctness of the acquired output relative to the true output, and the second index indicates a quality of the selected at least a part of the pieces of external knowledge;
and
learning a first neural network by updating parameters of the first neural network using the reward.

3 more elements

A computer-implemented method for learning, comprising:

determining scores based on obtained from similarity degrees between an input sentence and pieces of external knowledge stored in an external knowledge database;
selecting at least a part of the pieces of external knowledge based on the scores as search results;
acquiring an output to the input sentence based on predetermined arithmetic processing using the input sentence and the selected at least a part of the pieces of external knowledge as inputs;
determining, based on a combination of the input sentence, the acquired output, the selected pieces of external knowledge, and a true output given to the input sentence in advance, a reward, therein the reward is based on a first index and a second index, the first index indicates correctness of the output relative to the true output, and the second index indicates quality of the selected at least a part of the pieces of external knowledge;
and
learning a first neural network by updating parameters of the first neural network using the reward.

3 more elements

A computer-readable non-transitory recording medium storing a computer-executable program instructions that when executed by a processor cause a computer system to:

determine scores based on similarity degrees between an input sentence and pieces of external knowledge stored in an external knowledge database;
selecting at least a part of the pieces of external knowledge based on the scores as search results;
acquire an output to the input sentence based on predetermined arithmetic processing using the input sentence and the selected at least a part of the pieces of external knowledge as inputs;
determine based on a combination of the input sentence, the acquired output, the selected pieces of external knowledge, and a true output given to the input sentence in advance, a reward, wherein the reward is based on a first index and a second index, the first index indicates correctness of the output relative to the true output, and the second index indicates quality of the selected pieces of external knowledge;
and
learning a first neural network by updating parameters of the first neural network using the reward.

3 more elements

Dependencies are inferred from the claim text.

Learning device, learning method and learning program

Life of this patent

Learning device, learning method and learning program

Life of this patent

Abstract

Claims20