Unified representation learning of media features for diverse tasks

Abstract

Methods and systems are disclosed for generating general feature vectors (GFVs), each simultaneously constructed for separate tasks of image reconstruction and fingerprint-based image discrimination. The computing system may include machine-learning-based components configured for extracting GFVs from images, signal processing for both transmission and reception and recovery of the extracted GFVs, generating reconstructed images from the recovered GFVs, and discriminating between fingerprints generated from the recovered GFVs and query fingerprints generated from query GFVs. A set of training images may be received at the computing system. In each of one or more training iterations over the set of training images, the components may be jointly trained with each training image of the set by minimizing a joint loss function computed as a sum of losses due to signal processing and recovery, image reconstruction, and fingerprint discrimination. The trained components may be configured for runtime implementation among one or more computing devices.