High confidence predicted profiles for online dating matching using continuously learning machine learning models trained for each user with physical and facial attributes

Abstract

Machine learning models can be trained to vend profiles with a high likelihood of matching with a user comprises: training machine learning models with feature vectors representing potential profile matches identified as input to a user match selection process, updating the trained models with feature vectors identified as further input to the selection process, in response to the updating, swapping the trained model out of and a further trained machine learning model into a foreground execution space, resulting in the untrained model being in the foreground and the trained model being in a processing background, after satisfying a defined criterion, injecting an outlier potential match entity as supplemental input to the further training with a supplemental feature vector as further input to the selection process, determining respective confidence values corresponding to candidate profiles accessible to the selection process, and rendering profile images of candidate profiles for the selection process.