Brain activity is usually studied at very different scales. At one end, there are detailed neuron and synapse models. At the other, there are whole-brain signals such as EEG, MEG, and fMRI. But a lot of interesting neuroscience lives in the middle: local populations of excitatory and inhibitory cells, interacting across cortical circuits, producing activity that can later be observed at the scalp or in imaging data.
NOEMA is built for that middle ground.
The project explores a mesoscopic brain simulation framework: not a single-neuron simulator, and not only a coarse whole-brain model, but a system for modelling local cortical modules and connecting them into larger networks. Each module represents local excitatory/inhibitory population dynamics. These modules can then be linked by structural connectivity, simulated over time, and projected into simplified observation models that resemble EEG, MEG, or BOLD-like signals.
The core idea is to treat the mesoscopic module as the main unit of simulation. A module has its own local dynamics, but it is not isolated. It sends and receives signals from other modules, forming a coupled brain network. From there, the same underlying simulated activity can be viewed in multiple ways: as local population activity, as source or dipole-like activity, or as projected sensor-level signals.
That matters because many neuroscience questions are multiscale by nature. EEG and MEG do not directly record individual neurons; they reflect coordinated population activity. fMRI does not directly record spikes either; it reflects slower physiological consequences of neural dynamics. NOEMA is designed around this gap. It asks: how can local mesoscopic dynamics give rise to the macroscopic signals we actually measure?
At its current conceptual center is a familiar excitatory/inhibitory population model: Wilson-Cowan-style dynamics. This gives NOEMA a working starting point for simulating how local excitation and inhibition evolve over time. The broader ambition is to extend this toward richer finite-size population models, where the fact that a population contains a finite number of units is not just ignored but becomes part of the dynamics through structured stochasticity and mesoscopic variability.
The long-term vision is not only to simulate activity, but to make the simulation interpretable across levels. A researcher should be able to define a network of cortical modules, run the dynamics, inspect local population states, derive source-level activity, project it to EEG/MEG-like sensors, and compare simplified BOLD-like summaries. In that sense, NOEMA is less a single model than a scaffold for asking multiscale questions.
Visualization is part of the scientific idea, not just decoration. If the model is a network of interacting modules, then researchers need to see that structure: which modules exist, how they are connected, how activity moves through them, and how local dynamics relate to macroscopic observations. NOEMA therefore points toward both inline plotting and richer graphical interfaces for exploring model structure, dynamics, diagnostics, and multiscale relationships.
The broader ambition is to create a bridge between three levels: local population dynamics, where excitation, inhibition, and finite-size effects shape activity; network coupling, where modules interact through structural connections; and observation models, where simulated activity becomes something like EEG, MEG, or fMRI.