CADRE is a family of video autoencoders from Kyutai built for action-conditioned world models. A world model generates in the autoencoder’s latent space. So the autoencoder’s compression sets both ...
A new hybrid framework combining a Variational Autoencoder with Proximal Policy Optimization reinforcement learning generates ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results