DP-GP-LVM: A Bayesian Non-Parametric Model for Learning Multivariate Dependency Structures
This work addresses a domain-specific challenge in machine learning for handling complex dependencies in multivariate data, representing an incremental advancement in latent variable modeling.
The authors tackled the problem of learning multivariate dependency structures by introducing DP-GP-LVM, a non-parametric Bayesian latent variable model that uses Gaussian process and Dirichlet process priors, resulting in an efficient variational inference method with improved performance over previous models.
We present a non-parametric Bayesian latent variable model capable of learning dependency structures across dimensions in a multivariate setting. Our approach is based on flexible Gaussian process priors for the generative mappings and interchangeable Dirichlet process priors to learn the structure. The introduction of the Dirichlet process as a specific structural prior allows our model to circumvent issues associated with previous Gaussian process latent variable models. Inference is performed by deriving an efficient variational bound on the marginal log-likelihood on the model.