Learning high-dimensional mixed models via amortized variational inference

Priscilla Ong, Manuel Haussmann, Harri Lähdesmäki

Keywords: Latent Variable Model, Linear Mixed Model, Gaussian Processes, Longitudinal Data

TL;DR: We leverage linear mixed models and amortized variational inference to provide conditional priors for VAEs, and propose LMM-VAE, a scalable and interpretable model for modelling longitudinal data.

Abstract: Modelling longitudinal data is an important yet challenging task. These datasets can be high-dimensional, consist of non-linear effects, and contain time-varying covariates. In this work, we leverage linear mixed models (LMMs) and amortized variational inference to provide conditional priors for VAEs, and propose LMM-VAE, a model that is scalable, interpretable, and shares theoretical connections to the GP-based VAEs. We empirically demonstrate that LMM-VAE performs competitively compared to existing approaches.

Submission Number: 12