HINT: Hierarchical Invertible Neural Transport for Density Estimation and Bayesian Inference
This work addresses a bottleneck in invertible neural networks for density estimation and Bayesian inference, offering an incremental improvement in expressiveness for researchers in probabilistic machine learning.
The paper tackles the limited expressiveness of invertible neural networks with sparse Jacobians by introducing HINT, a hierarchical recursive coupling scheme that yields dense triangular Jacobians, enabling efficient sampling from joint and posterior distributions; it demonstrates competitive performance on standard datasets and introduces a novel 2D shapes dataset for consistent visualization.
Many recent invertible neural architectures are based on coupling block designs where variables are divided in two subsets which serve as inputs of an easily invertible (usually affine) triangular transformation. While such a transformation is invertible, its Jacobian is very sparse and thus may lack expressiveness. This work presents a simple remedy by noting that subdivision and (affine) coupling can be repeated recursively within the resulting subsets, leading to an efficiently invertible block with dense, triangular Jacobian. By formulating our recursive coupling scheme via a hierarchical architecture, HINT allows sampling from a joint distribution p(y,x) and the corresponding posterior p(x|y) using a single invertible network. We evaluate our method on some standard data sets and benchmark its full power for density estimation and Bayesian inference on a novel data set of 2D shapes in Fourier parameterization, which enables consistent visualization of samples for different dimensionalities.