RNADE: The real-valued neural autoregressive density-estimator
This work addresses density estimation for heterogeneous and perceptual data, representing an incremental improvement over existing mixture models.
The authors tackled joint density estimation of real-valued vectors by introducing RNADE, a model that calculates density as a product of one-dimensional conditionals using mixture density networks with shared parameters, and found it outperforms mixture models on most datasets.
We introduce RNADE, a new model for joint density estimation of real-valued vectors. Our model calculates the density of a datapoint as the product of one-dimensional conditionals modeled using mixture density networks with shared parameters. RNADE learns a distributed representation of the data, while having a tractable expression for the calculation of densities. A tractable likelihood allows direct comparison with other methods and training by standard gradient-based optimizers. We compare the performance of RNADE on several datasets of heterogeneous and perceptual data, finding it outperforms mixture models in all but one case.