Advances in Density Estimation and Generative Modeling

The field of density estimation and generative modeling is witnessing significant developments, with a focus on improving the efficiency, flexibility, and robustness of existing methods. Researchers are exploring new approaches to model complex distributions, including the use of kernelized matrix costs, random projection flows, and marginal flows. These innovations aim to address the limitations of current methods, such as expensive training, slow inference, and mode collapse. Noteworthy papers in this area include: Marginal Flow, which proposes a flexible and efficient framework for density estimation that overcomes the limitations of current approaches. Simple, Fast and Efficient Injective Manifold Density Estimation with Random Projections, which introduces a principled framework for injective normalizing flows that leverages tools from random matrix theory and the geometry of random projections. Efficient Probabilistic Tensor Networks, which proposes a conceptually simple approach for learning probabilistic tensor networks efficiently, achieving significant improvements in time and space complexity.

Sources

A Family of Kernelized Matrix Costs for Multiple-Output Mixture Neural Networks

Simple, Fast and Efficient Injective Manifold Density Estimation with Random Projections

Sampling via Gaussian Mixture Approximations

Marginal Flow: a flexible and efficient framework for density estimation

Efficient Probabilistic Tensor Networks

Adaptive Heterogeneous Mixtures of Normalising Flows for Robust Variational Inference

Built with on top of