Neuroprobe
Benchmark for evaluating decoding capabilities of iEEG foundation models.
| Rank | Model | Author | Organization | Date | Overall AUROC | Sentence Onset | Speech | Volume | Delta Volume | Voice Pitch | Word Position | Inter-word Gap | GPT-2 Surprisal | Head Word Position | Part of Speech | Word Length | Global Optical Flow | Local Optical Flow | Frame Brightness | Number of Faces |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | DIVER-1_0.1s_tiny_frozen | Yonghyeon Gwon | Seoul National University | 2026-01-09 | 0.678 | 0.930 | 0.897 | 0.719 | 0.803 | 0.592 | 0.798 | 0.636 | 0.629 | 0.623 | 0.627 | 0.646 | 0.628 | 0.616 | 0.495 | 0.525 |
| 2 | Population Transformer (global z-scoring) | Geeling Chau | Caltech | 2026-04-28 | 0.670 | 0.930 | 0.919 | 0.766 | 0.781 | 0.596 | 0.758 | 0.620 | 0.603 | 0.601 | 0.604 | 0.622 | 0.621 | 0.618 | 0.496 | 0.517 |
| 3 | CNN_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.669 | 0.886 | 0.888 | 0.757 | 0.762 | 0.618 | 0.734 | 0.583 | 0.597 | 0.614 | 0.621 | 0.636 | 0.638 | 0.630 | 0.536 | 0.530 |
| 4 | DIVER-1_0.1s_tiny | Yonghyeon Gwon | Seoul National University | 2026-01-09 | 0.665 | 0.924 | 0.897 | 0.688 | 0.814 | 0.561 | 0.799 | 0.634 | 0.618 | 0.617 | 0.607 | 0.638 | 0.594 | 0.584 | 0.486 | 0.518 |
| 5 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.660 | 0.888 | 0.882 | 0.717 | 0.756 | 0.584 | 0.740 | 0.613 | 0.608 | 0.603 | 0.601 | 0.617 | 0.630 | 0.615 | 0.528 | 0.519 |
| 6 | MLP_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.656 | 0.884 | 0.877 | 0.717 | 0.747 | 0.585 | 0.728 | 0.597 | 0.604 | 0.602 | 0.597 | 0.615 | 0.629 | 0.617 | 0.524 | 0.521 |
| 7 | BrainBERT (frozen, global z-scoring) | Geeling Chau | Caltech | 2026-05-02 | 0.626 | 0.883 | 0.887 | 0.723 | 0.677 | 0.564 | 0.620 | 0.526 | 0.557 | 0.567 | 0.576 | 0.586 | 0.609 | 0.599 | 0.497 | 0.516 |
| 8 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.625 | 0.837 | 0.816 | 0.719 | 0.711 | 0.566 | 0.651 | 0.582 | 0.567 | 0.564 | 0.557 | 0.565 | 0.604 | 0.596 | 0.522 | 0.522 |
| 9 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.604 | 0.792 | 0.645 | 0.597 | 0.742 | 0.535 | 0.742 | 0.598 | 0.575 | 0.569 | 0.579 | 0.595 | 0.536 | 0.541 | 0.507 | 0.503 |
| 10 | BrainBERT (frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.585 | 0.754 | 0.606 | 0.576 | 0.699 | 0.527 | 0.679 | 0.581 | 0.580 | 0.582 | 0.556 | 0.571 | 0.527 | 0.520 | 0.510 | 0.509 |
| 11 | BrainBERT (untrained; frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.581 | 0.749 | 0.598 | 0.566 | 0.690 | 0.519 | 0.677 | 0.581 | 0.579 | 0.581 | 0.557 | 0.569 | 0.525 | 0.520 | 0.495 | 0.506 |
| 12 | GLIS-GNN (ST-GCN, Functional Graph) | Eli Gross | Columbia University | 2026-04-14 | 0.553 | 0.710 | 0.708 | 0.601 | 0.581 | 0.516 | 0.534 | 0.505 | 0.522 | 0.523 | 0.521 | 0.520 | 0.524 | 0.537 | 0.472 | 0.515 |
| 13 | PopulationTransformer (off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.549 | 0.741 | 0.672 | 0.562 | 0.622 | 0.527 | 0.528 | 0.523 | 0.517 | 0.516 | 0.507 | 0.521 | 0.515 | 0.501 | 0.489 | 0.500 |
| Rank | Model | Author | Organization | Date | Overall AUROC | Sentence Onset | Speech | Volume | Delta Volume | Voice Pitch | Word Position | Inter-word Gap | GPT-2 Surprisal | Head Word Position | Part of Speech | Word Length | Global Optical Flow | Local Optical Flow | Frame Brightness | Number of Faces |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | CNN_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.670 | 0.919 | 0.919 | 0.731 | 0.753 | 0.618 | 0.712 | 0.581 | 0.622 | 0.619 | 0.638 | 0.638 | 0.629 | 0.608 | 0.524 | 0.546 |
| 2 | Population Transformer (global z-scoring) | Geeling Chau | Caltech | 2026-04-28 | 0.663 | 0.924 | 0.913 | 0.712 | 0.759 | 0.598 | 0.703 | 0.602 | 0.624 | 0.612 | 0.611 | 0.636 | 0.607 | 0.594 | 0.505 | 0.541 |
| 3 | MLP_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.655 | 0.910 | 0.902 | 0.705 | 0.745 | 0.587 | 0.688 | 0.578 | 0.609 | 0.603 | 0.613 | 0.611 | 0.617 | 0.592 | 0.528 | 0.536 |
| 4 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.651 | 0.913 | 0.908 | 0.708 | 0.737 | 0.588 | 0.696 | 0.584 | 0.596 | 0.585 | 0.612 | 0.608 | 0.598 | 0.580 | 0.532 | 0.522 |
| 5 | BrainBERT (frozen, global z-scoring) | Geeling Chau | Caltech | 2026-05-02 | 0.633 | 0.905 | 0.903 | 0.705 | 0.674 | 0.584 | 0.612 | 0.526 | 0.573 | 0.567 | 0.592 | 0.600 | 0.604 | 0.590 | 0.522 | 0.532 |
| 6 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.626 | 0.862 | 0.860 | 0.712 | 0.697 | 0.563 | 0.649 | 0.556 | 0.564 | 0.555 | 0.559 | 0.566 | 0.589 | 0.587 | 0.545 | 0.521 |
| 7 | BrainBERT (frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.580 | 0.737 | 0.632 | 0.570 | 0.684 | 0.513 | 0.658 | 0.566 | 0.578 | 0.577 | 0.552 | 0.559 | 0.526 | 0.537 | 0.509 | 0.502 |
| 8 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.574 | 0.724 | 0.616 | 0.565 | 0.699 | 0.525 | 0.654 | 0.544 | 0.559 | 0.538 | 0.564 | 0.554 | 0.527 | 0.521 | 0.507 | 0.510 |
| 9 | BrainBERT (untrained; frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.572 | 0.724 | 0.609 | 0.557 | 0.671 | 0.506 | 0.658 | 0.560 | 0.576 | 0.578 | 0.554 | 0.558 | 0.522 | 0.525 | 0.493 | 0.496 |
| 10 | PopulationTransformer (off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.564 | 0.754 | 0.719 | 0.572 | 0.637 | 0.523 | 0.568 | 0.526 | 0.554 | 0.535 | 0.503 | 0.522 | 0.516 | 0.518 | 0.496 | 0.513 |
| 11 | GLIS-GNN (ST-GCN, Functional Graph) | Eli Gross | Columbia University | 2026-04-14 | 0.550 | 0.679 | 0.716 | 0.555 | 0.563 | 0.527 | 0.532 | 0.496 | 0.535 | 0.528 | 0.530 | 0.513 | 0.547 | 0.505 | 0.513 | 0.510 |
| 12 | RNN (gru) | Bennett Stankovits | MIT | 2026-02-02 | 0.510 | 0.531 | 0.537 | 0.512 | 0.518 | 0.498 | 0.507 | 0.494 | 0.514 | 0.511 | 0.514 | 0.509 | 0.503 | 0.506 | 0.493 | 0.503 |
| Rank | Model | Author | Organization | Date | Overall AUROC | Sentence Onset | Speech | Volume | Delta Volume | Voice Pitch | Word Position | Inter-word Gap | GPT-2 Surprisal | Head Word Position | Part of Speech | Word Length | Global Optical Flow | Local Optical Flow | Frame Brightness | Number of Faces |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | CNN_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.578 | 0.780 | 0.751 | 0.587 | 0.662 | 0.516 | 0.617 | 0.545 | 0.539 | 0.555 | 0.520 | 0.511 | 0.531 | 0.528 | 0.516 | 0.505 |
| 2 | Population Transformer (global z-scoring) | Geeling Chau | Caltech | 2026-04-28 | 0.575 | 0.711 | 0.714 | 0.589 | 0.653 | 0.525 | 0.630 | 0.525 | 0.563 | 0.553 | 0.523 | 0.527 | 0.548 | 0.547 | 0.502 | 0.516 |
| 3 | MLP_Laplacian_rereferencing_spectrogram | Geeling Chau | Caltech | 2026-04-21 | 0.566 | 0.749 | 0.729 | 0.571 | 0.634 | 0.514 | 0.602 | 0.530 | 0.532 | 0.548 | 0.517 | 0.521 | 0.523 | 0.512 | 0.502 | 0.502 |
| 4 | BrainBERT (frozen, global z-scoring) | Geeling Chau | Caltech | 2026-05-02 | 0.547 | 0.712 | 0.722 | 0.536 | 0.564 | 0.529 | 0.539 | 0.499 | 0.507 | 0.540 | 0.515 | 0.507 | 0.534 | 0.514 | 0.497 | 0.492 |
| 5 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.539 | 0.676 | 0.643 | 0.534 | 0.569 | 0.505 | 0.566 | 0.515 | 0.508 | 0.525 | 0.506 | 0.511 | 0.515 | 0.510 | 0.495 | 0.511 |
| 6 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.528 | 0.616 | 0.575 | 0.540 | 0.548 | 0.503 | 0.558 | 0.509 | 0.510 | 0.511 | 0.509 | 0.512 | 0.509 | 0.501 | 0.519 | 0.494 |
| 7 | BrainBERT (untrained; frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.527 | 0.583 | 0.540 | 0.521 | 0.589 | 0.506 | 0.570 | 0.517 | 0.516 | 0.531 | 0.516 | 0.510 | 0.503 | 0.499 | 0.501 | 0.498 |
| 8 | BrainBERT (frozen; off-the-shelf; per-window STFT z-scoring) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.522 | 0.584 | 0.537 | 0.526 | 0.571 | 0.498 | 0.549 | 0.510 | 0.515 | 0.527 | 0.505 | 0.509 | 0.501 | 0.497 | 0.509 | 0.498 |
| 9 | GLIS-GNN (ST-GCN, Functional Graph) | Eli Gross | Columbia University | 2026-04-14 | 0.515 | 0.608 | 0.541 | 0.559 | 0.519 | 0.490 | 0.503 | 0.498 | 0.508 | 0.497 | 0.500 | 0.488 | 0.483 | 0.538 | 0.497 | 0.498 |
| 10 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2026-05-02 | 0.509 | 0.534 | 0.504 | 0.517 | 0.529 | 0.501 | 0.536 | 0.510 | 0.506 | 0.506 | 0.496 | 0.504 | 0.500 | 0.498 | 0.496 | 0.501 |
Neuroprobe: Evaluating Intracranial Brain Responses to Naturalistic Stimuli
¹MIT CSAIL, CBMM | ²MIT McGovern Institute | ³Caltech | *Equal contribution
Citation
If you use Neuroprobe in your work, please cite our paper:
@misc{neuroprobe,
title={Neuroprobe: Evaluating Intracranial Brain Responses to Naturalistic Stimuli},
author={Andrii Zahorodnii and Christopher Wang and Bennett Stankovits and Charikleia Moraitaki and Geeling Chau and Andrei Barbu and Boris Katz and Ila R Fiete},
year={2025},
eprint={2509.21671},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2509.21671},
}