A Deep and Tractable Density Estimator
About
The Neural Autoregressive Distribution Estimator (NADE) and its real-valued version RNADE are competitive density models of multidimensional data across a variety of domains. These models use a fixed, arbitrary ordering of the data dimensions. One can easily condition on variables at the beginning of the ordering, and marginalize out variables at the end of the ordering, however other inference tasks require approximate inference. In this work we introduce an efficient procedure to simultaneously train a NADE model for each possible ordering of the variables, by sharing parameters across all these models. We can thus use the most convenient model for each inference task at hand, and ensembles of such models with different orderings are immediately available. Moreover, unlike the original NADE, our training procedure scales to deep models. Empirically, ensembles of Deep NADE models obtain state of the art density estimation performance.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Density Estimation | binarized MNIST 28x28 (test) | Test LogL-84.55 | 44 | |
| Density Estimation | Ocr-letters (test) | Avg Log-Likelihood (nats)-27.22 | 19 | |
| Density Estimation | UCI Red wine (test) | Avg Test Log-Likelihood-8.76 | 18 | |
| Density Estimation | UCI White wine (test) | Average Test Log-Likelihood-9.67 | 18 | |
| Density Estimation | UCI Parkinsons (test) | Avg Test Log-Likelihood-0.9 | 18 | |
| Generative Modeling | MNIST Binary (test) | NLL (nats)85.1 | 13 | |
| Density Estimation | BSDS300 8x8 pixel patches (test) | Avg Test-Set Log-Likelihood157 | 10 | |
| Density Estimation | Connect4 (test) | Avg Log-Likelihood (nats)-11.99 | 9 | |
| Density Estimation | dna (test) | Avg Log-Likelihood (nats)-82.31 | 9 | |
| Density Estimation | Mushrooms (test) | Avg Log-Likelihood (nats)-9.68 | 9 |