Training apparatus, training method, and storage medium

A training method comprising:

• inputting a bag that is a set of an instance to an identifier and outputting a probability sequence including a probability that each instance belonging to the bag corresponds to a detection target;
• calculating a bag likelihood based on the probability sequence, the bag likelihood being a probability that one or more specific instances corresponding to the detection target are included in the bag;
• outputting an expected instance quantity based on the probability sequence, the expected instance quantity being a numerical value regarding a quantity of the specific instances included in the bag;
• and
• calculating a parameter update amount that is an amount to which a parameter of the identifier is updated based on teaching data that is a numerical value regarding a quantity of the specific instances included in the bag, the bag likelihood, and the expected instance quantity, and updating a parameter of the identifier based on the parameter update amount.

A non-transitory computer readable medium including computer executable instructions, wherein the instructions, when executed by a processor, cause the processor to perform operations comprising:

• inputting a bag that is a set of an instance to an identifier and outputting a probability sequence including a probability that each instance belonging to the bag corresponds to a detection target;
• calculating a bag likelihood based on the probability sequence, the bag likelihood being a probability that one or more specific instances corresponding to the detection target are included in the bag;
• outputting an expected instance quantity based on the probability sequence, the expected instance quantity being a numerical value regarding a quantity of the specific instances included in the bag;
• and
• calculating a parameter update amount that is an amount to which a parameter of the identifier is updated based on teaching data that is a numerical value regarding a quantity of the specific instances included in the bag, the bag likelihood, and the expected instance quantity, and updating a parameter of the identifier based on the parameter update amount.