TY  - JOUR
AU  - Perrakis, Konstantinos
AU  - Lartigue, Thomas-Alan-Jean
AU  - Dondelinger, Frank
AU  - Mukherjee, Sach
TI  - Regularized Joint Mixture Models
JO  - Journal of machine learning research
VL  - 24
IS  - 19
SN  - 1532-4435
CY  - Brookline, MA
PB  - Microtome Publishing
M1  - DZNE-2025-00171
SP  - 1 - 47
PY  - 2023
AB  - Regularized regression models are well studied and, under appropriate conditions, offerfast and statistically interpretable results. However, large data in many applications areheterogeneous in the sense of harboring distributional differences between latent groups.Then, the assumption that the conditional distribution of response Y given features X is thesame for all samples may not hold. Furthermore, in scientific applications, the covariancestructure of the features may contain important signals and its learning is also affected bylatent group structure. We propose a class of mixture models for paired data pX, Y q thatcouples together the distribution of X (using sparse graphical models) and the conditionalY  -  X (using sparse regression models). The regression and graphical models are specificto the latent groups and model parameters are estimated jointly. This allows signals ineither or both of the feature distribution and regression model to inform learning of latentstructure and provides automatic control of confounding by such structure. Estimationis handled via an expectation-maximization algorithm, whose convergence is establishedtheoretically. We illustrate the key ideas via empirical examples. An R package is availableat https://github.com/k-perrakis/regjmix.
LB  - PUB:(DE-HGF)16
UR  - https://pub.dzne.de/record/276090
ER  -