Explore how a brain–computer interface could adapt to a changing brain.
An interactive lab and reproducible research workspace for entropy-informed feedback.
Launch locally · Read the paper · Open the notebook · Explore the presentations · Hosted preview
Brain signals change with context. Could a feedback system use both their activity and their predictability to make better adjustments—and know when its measurements stop being useful?
This project investigates that question in the context of future, person-specific autism neuromodulation research. It is a synthetic research model, not a demonstrated therapy or a model of a diagnosed autistic person.
Actual model output rendered on template anatomy. Illustrative populations and connections; no patient data.
| Explore | What you can do |
|---|---|
| A brain in the loop | Rotate the 3D cortex, inspect populations, and follow activity over time. |
| Feedback that adapts | Compare six strategies, from no input to rate-plus-entropy control. |
| Conditions that matter | Add measurement noise, delay, or an optional plasticity rule. See what breaks. |
| Evidence you can inspect | Trace the equations, run the Python model, and examine the recorded results. |
24 synthetic individuals · 1,296 evaluation runs · 22 cited papers
The central result is conditional: adding an entropy term to rate feedback improved tracking under clean measurements, but noise reversed that advantage. Task accuracy did not recover. Those limits are part of the finding—and a starting point for better experiments. See the results →
With Git and Python 3 installed:
git clone https://github.com/ProCO-RPHopkins/BBCI-Adaptive-Dynamics-Lab.git
cd BBCI-Adaptive-Dynamics-Lab
python -m http.server 8000 --bind 127.0.0.1 --directory lab/distOpen localhost:8000. No npm installation or build step. The viewer, cortical geometry, and reference data are included. Use a browser with WebGL support.
View the hosted preview.
Run the Python benchmark or notebook
From the repository root, using Python 3.12:
python -m venv .venv
source .venv/bin/activate
# Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install -r research/requirements.txt
cd research
jupyter lab BBCI_Research_Notebook.ipynbTo reproduce the full study, run python run_experiments.py, then python validate.py and python make_figures.py from research/. These commands update the recorded outputs. The notebook runs its demonstration and displays the recorded aggregate study; it does not rerun all 1,296 evaluations.
See the research guide for calibration, dependencies, execution details, and limitations.
- The science: manuscript PDF, editable Word file, and evidence ledger.
- The story: research explained and research opportunity—both with full speaker scripts and citations.
- The machinery: Python model, browser source, results and figures, and artifact authoring.
- The journey: earlier work and the Julia prototype audit.
Useful next contributions include realistic observation models, stronger comparison controllers, independent replication, and outcome design with autistic participants. Start a discussion in an issue with a question, reproducible finding, or proposed experiment.
Research by Ryan Hopkins, with disclosed AI assistance. The manuscript is a submission draft awaiting independent review. Cite this project · Provenance · Submission notes
Original code: MIT. Original research content: CC BY 4.0. Third-party materials retain their own terms; see rights and attribution.