Hacker News new | ask | show | jobs
by danlenton 747 days ago
We just initialize a random latent vector for each model, and then jointly train each of these unique latent vectors :)