Deep probabilistic model synthesis enables unified modeling of whole-brain neural activity across individual subjects

2026-03-14

AI Summary

Traditional machine learning models struggle to effectively integrate data from multiple homogeneous systems—such as individual brains—to construct a unified model that captures both shared patterns and system-specific characteristics. To address this challenge, this work proposes the Deep Probabilistic Model Synthesis (DPMS) framework, which introduces system-level auxiliary attributes into the model synthesis process for the first time. By leveraging variational inference, DPMS jointly learns a conditional prior encoding commonalities across systems and instance-specific posteriors capturing individual variations, thereby providing a unified probabilistic treatment of both shared structure and idiosyncratic features. The method is applicable to regression, classification, and dimensionality reduction tasks, and demonstrates significant performance gains over single-instance models on both synthetic benchmarks and whole-brain neural activity recordings from juvenile zebrafish, enabling efficient cross-individual probabilistic modeling and integration of neural dynamics.

Technology Category

• Machine Learning: Probabilistic Circuits and Graphical Models
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Many disciplines need quantitative models that synthesize experimental data across multiple instances of the same general system. For example, neuroscientists must combine data from the brains of many individual animals to understand the species' brain in general. However, typical machine learning models treat one system instance at a time. Here we introduce a machine learning framework, deep probabilistic model synthesis (DPMS), that leverages system properties auxiliary to the model to combine data across system instances. DPMS specifically uses variational inference to learn a conditional prior distribution and instance-specific posterior distributions over model parameters that respectively tie together the system instances and capture their unique structure. DPMS can synthesize a wide variety of model classes, such as those for regression, classification, and dimensionality reduction, and we demonstrate its ability to improve upon single-instance models on synthetic data and whole-brain neural activity data from larval zebrafish.

Problem

• model synthesis
• cross-subject modeling
• whole-brain neural activity
• probabilistic modeling
• data integration

Innovation

• deep probabilistic model synthesis
• variational inference
• cross-instance modeling
• conditional prior
• whole-brain neural activity