A new machine learning framework called Deep Probabilistic Model Synthesis (DPMS) can combine brain-activity data from different individuals into a single unified model. This approach, developed by William Bishop and colleagues at Janelia Research Campus and Radboud University, outperforms models trained on just one animal at a time.
The research
Published on arXiv (June 2026), the team used DPMS to analyze whole-brain neural activity from larval zebrafish. Typical ML models are designed for one instance—for example, one brain—but DPMS uses a conditional prior and instance-specific posteriors to account for both shared and unique features across subjects. They tested the framework on synthetic data and on a dataset of around 100,000 neurons from multiple zebrafish. The DPMS models consistently produced better reconstructions of neural activity than single-subject models, with improvements of up to 20% in predictive accuracy on held-out data. The framework works with regression, classification, and dimensionality reduction models.
Why it matters
The ability to pool neural data across individuals—without losing each brain's unique wiring—could accelerate our understanding of general brain principles. For anyone curious about cognition, this means future brain-training tools or cognitive assessments could be validated on more robust models that represent the average brain while still respecting personal differences. This fits well with IQ testing and brain training, where personalized insights are key.
What you can do
While DPMS is a research tool, you can support better brain science by participating in citizen science projects that collect cognitive data. Meanwhile, take scientifically validated IQ tests and brain training exercises that adapt to your performance—they mirror the personalized approach DPMS enables.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
