| CPC G06N 20/00 (2019.01) [G06N 5/02 (2013.01)] | 11 Claims |
|
1. A training apparatus comprising:
a processor programmed to:
acquire a first model including:
an input layer to which input information is input;
a plurality of intermediate layers that executes a calculation based on a feature of the input information that has been input; and
an output layer that outputs output information that corresponds to output of the intermediate layer;
acquire a second model that has learnt a feature of third information; and
train the first model such that:
when predetermined input information is input to the first model, the first model outputs predetermined output information that corresponds to the predetermined input information and intermediate information output from a predetermined intermediate layer among the intermediate layers becomes closer to feature information that corresponds to a feature of correspondence information that corresponds to the predetermined input information;
when input information related to a first domain is input as the predetermined input information to the first model, information indicating classification of the input information is output as the output information and the intermediate information becomes closer to feature information that takes account of correspondence information related to a second domain different from the first domain; and
when first information and second information associated with the first information are input as the predetermined input information to the first model, a classification result of the second information is output as the output information and the intermediate information becomes closer to feature information that corresponds to a feature of the second information and that takes account of a feature of the third information associated with the first information,
the intermediate information is a vector output from a hidden layer of a classifier that is part of the first model at a stage prior to an output layer of the classifier into a function describing a term for training the first model and so that the intermediate information becomes closer to feature information generated from the second information by the second model, the feature information being a vector output from a predetermined hidden intermediate layer among a plurality of intermediate layers included in the second model and input into the function describing the term, the second model being an autoencoder and the predetermined hidden intermediate layer being an intermediate layer prior to decoding,
wherein closeness of information is determined by vector distance.
|