US patent US12802070
Offline data transfer utilizing machine learning models
Abstract
Techniques described herein are directed to receiving an embedding indicating first characteristics of a user profile. That embedding is exchanged with another device, which also sends another embedding associated with that device. An on-device machine learning model is utilized to analyze the embedding along with an embedding of the transaction at issue to determine a risk metric associated with the transaction. The transaction can then be effectuated if the users agree to proceed with the transaction given the risk metric.