/research/bci

Brain–computer interfaces

We work on the signal chain behind non-invasive BCIs: acquiring EEG, cleaning it, extracting what matters and classifying it in real time.

01 / Scope

What we work on.

01
Signal acquisition and quality
Electrode contact, sampling and referencing choices, and continuous signal-quality monitoring during a session.
02
Preprocessing
Filtering, line-noise removal and handling of eye and muscle artifacts before any model sees the data.
03
Neural data analysis
Time–frequency analysis, event-related responses and feature extraction from multichannel recordings.
04
Real-time classification
Low-latency pipelines that classify brain states on streaming data, not only on recorded datasets.

02 / Approach

Step 01
Start from the signal
Many BCI problems are data-quality problems. We look at the raw signal before we look at accuracy.
Step 02
Offline, then online
Methods are validated on recorded sessions first, then moved to streaming pipelines with a fixed latency budget.
Step 03
Simple, inspectable models first
We prefer models whose decisions we can examine, and add complexity only when the data justifies it.
Step 04
Write it down
Methods, failures and open questions go into Lab Notes, including our open BCI lecture notes.

We are open to research partnerships with universities, labs and clinical teams, from shared datasets to joint projects.