Reweighted Wake-Sleep
This work addresses the problem of efficient training and inference in deep generative models for machine learning researchers, offering incremental improvements over existing wake-sleep methods.
The paper tackles the challenge of training deep directed graphical models with many hidden variables by proposing a reweighted wake-sleep algorithm that improves gradient estimators through multiple sampling from the inference network, achieving better likelihood in experiments. It also shows that using more powerful models like NADE for inference layers yields substantially better generative models.
Training deep directed graphical models with many hidden variables and performing inference remains a major challenge. Helmholtz machines and deep belief networks are such models, and the wake-sleep algorithm has been proposed to train them. The wake-sleep algorithm relies on training not just the directed generative model but also a conditional generative model (the inference network) that runs backward from visible to latent, estimating the posterior distribution of latent given visible. We propose a novel interpretation of the wake-sleep algorithm which suggests that better estimators of the gradient can be obtained by sampling latent variables multiple times from the inference network. This view is based on importance sampling as an estimator of the likelihood, with the approximate inference network as a proposal distribution. This interpretation is confirmed experimentally, showing that better likelihood can be achieved with this reweighted wake-sleep procedure. Based on this interpretation, we propose that a sigmoidal belief network is not sufficiently powerful for the layers of the inference network in order to recover a good estimator of the posterior distribution of latent variables. Our experiments show that using a more powerful layer model, such as NADE, yields substantially better generative models.