Zero-Calibration Attention Decoding by Cogitat
Abstract
The demo aims to introduce attendees to the Cogitat attention decoder, tracking real-time cognitive state while attendees engage with different tasks. It demonstrates zero-calibration decoding on a lightweight off-the-shelf headset, enabled by our EEG foundation model. This means insights are immediately available, without requiring lengthy calibration periods on different users. Unlike common band power metrics of attention, our model extracts the attention dynamics from the semantic space of brain activity in a self-guided way. This overcomes the limits of human-labeled data on poorly understood cognitive states like attention. The decoder is robust to common spurious features like eye movements and other contamination that often drives the predictions of other real-time EEG decoding systems.