Training device and training method that perform inference processing using a fusion DNN model

A training device, comprising:

• a processor;
• and
• a memory,
• wherein, by using a program recorded in the memory,
• the processor trains a deep neural network (DNN) model by (i) using training data that includes two or more modal information items and ground truth labels of the two or more modal information items and (ii) performing knowledge distillation that is a technique in which knowledge obtained as a result of a teacher model being trained is used to train a student model,
• the DNN model includes:
• two or more DNNs;
• and a fusion that includes a configuration in which portions of the two or more DNNs are fused and that receives an input of features that are outputs of the two or more DNNs,
• a first DNN, of the two or more DNNs, receives a first input of first modal information included in the two or more modal information items,
• the first DNN includes a Conv2D layer, a Global Average Pooling layer, and a Fully Connected layer,
• the Conv2D layer is a convolution layer that outputs two-dimensional feature maps obtained by performing convolution to extract features of the first modal information,
• the Global Average Pooling layer averages outputs of the two-dimensional feature maps output from the Conv2D layer and outputs a two-dimensional result, and
• the Fully Connected layer flattens the two-dimensional result output from the Global Average Pooling layer, by transforming the two-dimensional result into one-dimensional data, and outputs the one-dimensional data.