AI RESEARCH
Generalizing Linear Autoencoder Recommenders with Decoupled Expected Quadratic Loss
arXiv CS.LG
•
ArXi:2603.07402v1 Announce Type: new Linear autoencoders (LAEs) have gained increasing popularity in recommender systems due to their simplicity and strong empirical performance. Most LAE models, including the Emphasized Denoising Linear Autoencoder