Temporal Generative Models for Learning Heterogeneous Group Dynamics of Ecological Momentary Data.

Area of Science:

• Computational psychiatry
• Machine learning for mental health
• Time-series analysis of psychological data

Background:

• Precision psychiatry aims for individualized mental disorder characterization.
• Ecological Momentary Assessments (EMAs) provide high-frequency, real-time data but are complex (multi-dimensional, correlated, hierarchical).
• Existing models like mixed-effects models and standard Recurrent Temporal Restricted Boltzmann Machines (RTRBMs) have limitations in handling data heterogeneity and covariate incorporation.

Purpose of the Study:

• To propose a novel temporal generative model, the Heterogeneous-Dynamics Restricted Boltzmann Machine (HDRBM).
• To address the limitations of existing models in capturing heterogeneous group dynamics within populations using covariates.
• To demonstrate the effectiveness of HDRBM on simulated and real-world EMA data.

Main Methods:

• Development of the Heterogeneous-Dynamics Restricted Boltzmann Machine (HDRBM), a new temporal generative model.
• Application of HDRBM to analyze simulated and real-world Ecological Momentary Assessment (EMA) data.
• Incorporation of covariates within the generative model to account for population heterogeneity.

Main Results:

• HDRBM demonstrates improved accuracy and interpretability in modeling EMA data compared to existing approaches.
• The model successfully identifies underlying drivers of participant group dynamics.
• HDRBM serves as an effective generative model for complex, high-frequency psychological data.

Conclusions:

• The Heterogeneous-Dynamics Restricted Boltzmann Machine (HDRBM) offers a powerful new tool for precision psychiatry.
• Incorporating covariates into temporal generative models enhances the analysis of heterogeneous EMA data.
• HDRBM facilitates a deeper, individualized understanding of mental disorder dynamics.