JP-WO2025046718-A5 -

Dates

Publication Date

20260511

Application Date

20230829

Description

A learning device relating to an exemplary aspect of this disclosure comprises a calculation means for calculating features from training data including time-series image data, and a determination means for determining the parameters of a learning model so that the evaluation value for the difference in features over a period of time decreases. A learning method relating to an exemplary aspect of this disclosure includes a computation process in which at least one processor calculates features from training data including time-series image data, and a determination process in which the at least one processor determines the parameters of a learning model such that the evaluation value for the difference in features over a period of time decreases. An exemplary aspect of this disclosure is a program that causes a computer to function as a learning device, wherein the computer functions as a calculation means for calculating features from training data including time-series image data, and a determination means for determining the parameters of a learning model so that the evaluation value for the difference in features over a period of time decreases. <Configuration of the learning device> The configuration of the learning device 1 will be described with reference to Figure 1. Figure 1 is a block diagram showing the configuration of the learning device 1. As shown in Figure 1, the learning device 1 comprises a calculation unit 11 and a determination unit 12. The calculation unit 11 calculates features from training data, which includes time-series image data. The determination unit 12 determines the parameters of the learning model so that the evaluation value related to the difference in features over a period of time decreases. <Effects of learning devices> As described above, the learning device 1 employs a configuration that includes a calculation unit 11 that calculates features from training data including time-series image data, and a determination unit 12 that determines the parameters of the learning model so that the evaluation value related to the difference in features over a period of time decreases. Therefore, the learning device 1 has the effect of stabilizing the learning of a learning model using features. <Learning Method Flow> The flow of the learning method S1 will be explained with reference to Figure 2. Figure 2 is a flowchart showing the flow of the learning method S1. As shown in Figure 2, the learning method S1 includes a calculation process S11 and a decision process S12. In the calculation process S11, at least one processor calculates features from training data, which includes time-series image data. In the decision process S12, at least one processor determines the parameters of the learning model so that the evaluation value regarding the difference in features over a period of time decreases. <Effectiveness of learning methods> As described above, the learning method S1 employs a configuration that includes a calculation process in which at least one processor calculates features from training data including time-series image data, and a determination process in which at least one processor determines the parameters of the learning model so that the evaluation value regarding the difference in features over a period of time decreases. Therefore, the learning method S1 has the effect of stabilizing the learning of the learning model using features. The evaluation value, as an example, represents the change between the difference in features during the first period and the difference in features during a second period that is different from the first period. In this case, a small evaluation value can be said to mean that the time change of the difference in features before and after the transition is smooth. That is, in this case, the update unit 14A updates the model parameters so that the transition of features becomes smoother in at least one of the time series of features extracted by the feature extraction unit 12A and the time series of features predicted by the feature prediction unit 13A. Furthermore, the above evaluation value may, for example, represent the change between the difference in feature quantities in the first period and the difference in feature quantities in a second period that is different from the first period, and the above evaluation value may be larger the more similar the behavioral information in the first period and the behavioral information in the second period are. Also, for example, the above evaluation value may be larger the greater the discrepancy between the standard for the difference of each piece of behavioral information and the difference in the period. The regularization term calculation unit 163b adjusts at least one of the smoothness and sparsity according to the similarity between the actions before and after the transition. As an example, the regularization term calculation unit 163b calculates a regularization term that increases