EEG signal processing · Brain dynamics · Medical robotics · AI agents
Reading the oscillations of the brain, building robots for its vessels, and testing what AI agents actually do.
I work at Aicumene on multichannel EEG, from filtering and classifying it in real time to counting the brain’s active oscillators. My work also includes LevshAI, a robotic ecosystem for endovascular neurosurgery. With Alexandra Bernadotte I co-author Aicumene’s research on brain dynamics, strong AI, AI agents in economic games and verifiable AI for law.
Research
Most of this work starts from one question: how many oscillatory sources does a multichannel signal hold, and what are they? The same measurements then apply to the brain in sleep and coma, to trained networks, and the same habit of checking applies to AI agents.
Signal processing
Multichannel singular spectrum analysis embeds all channels at once, estimates how many oscillatory sources the signal holds, and recovers their frequencies, damping and spatial patterns, including attenuating oscillations.
Brain dynamics
BOID counts the active oscillatory modes in multichannel EEG. In coma after cardiac arrest it marks consciousness, a recovery mode and failed downshifting.
Learning systems
The same measurements applied to a trained artificial network: learning creates functional switching between oscillatory regimes.
EEG decoding
Silent speech recognised from EEG by convolutional and recurrent networks, with 85 % accuracy on nine words; and how the EEG of a simple speech task changes after COVID-19.
AI agents
Agents from different model families play classic economic games against each other. The work asks when their play converges to the Nash equilibrium.
Verifiable AI
A legal verifiability compiler with a firewall between rules and judgment, so that the conclusions of a legal AI system can be verified.
Engineering
The systems behind the research: signal pipelines that run in real time, robotics and imaging for the brain’s vessels, and the data and code released with the papers.
Medical robotics
A robotic ecosystem for endovascular operations: a motion-copying robot with haptic feedback, and a personalised 3D simulator of the patient’s vessels for planning before the operation.
Imaging
Semi-automatic annotation of brain vessels in magnetic resonance angiography images, published as a data paper.
Real-time EEG
Hankel embedding in a pipeline light enough to filter and classify EEG in real time.
Open data and code
The AiCumene EEG-BCI Sleep–Wake Dataset: 45 synthetic whole-night recordings that keep the oscillatory structure of their sources, public under CC BY-NC 4.0, and 40 real wearable-EEG recordings available on request. The MSSA and BOID libraries, and Aicumene’s open engineering tools.
Publications
Contact
Open to researchers, engineers and clinicians working on EEG and neurotechnology, medical robotics, and AI agents that can be checked.