US Patent 11631033 On-device training of personalized model

Methods related to a model training scheme performed on a computing device are described. The computing device may include a display module and a memory module storing multiple prediction models and item data that includes data of multiple items. Example methods may include displaying, on the display module, a first representation of at least one modality of one or more items among the multiple items; detecting an activity of a user of the computing device in response to displaying of the at least one modality of the one or more items to generate user activity log data; and evaluating at least one first prediction model of the multiple prediction models based on the user activity log data to generate an evaluation value of the at least one first prediction model. The at least one first prediction model may be related to a prediction of the one or more items.