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 | 2026-01-08 | 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 | DIVER-1_0.1s_tiny | Yonghyeon Gwon | Seoul National University | 2026-01-08 | 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 |
| 3 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.660 | 0.891 | 0.883 | 0.717 | 0.762 | 0.578 | 0.740 | 0.612 | 0.613 | 0.602 | 0.605 | 0.618 | 0.625 | 0.607 | 0.521 | 0.530 |
| 4 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.630 | 0.851 | 0.825 | 0.726 | 0.718 | 0.570 | 0.657 | 0.579 | 0.570 | 0.565 | 0.559 | 0.569 | 0.604 | 0.593 | 0.533 | 0.525 |
| 5 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.606 | 0.795 | 0.656 | 0.595 | 0.753 | 0.536 | 0.742 | 0.595 | 0.584 | 0.570 | 0.576 | 0.599 | 0.535 | 0.544 | 0.507 | 0.499 |
| 6 | BrainBERT (frozen; Wang et al. 2023) | - | MIT | 2025-09-25 | 0.586 | 0.757 | 0.611 | 0.583 | 0.706 | 0.524 | 0.685 | 0.584 | 0.580 | 0.585 | 0.556 | 0.571 | 0.521 | 0.525 | 0.508 | 0.503 |
| 7 | BrainBERT (untrained, frozen) | - | MIT | 2025-09-25 | 0.585 | 0.750 | 0.603 | 0.570 | 0.697 | 0.524 | 0.684 | 0.583 | 0.581 | 0.587 | 0.553 | 0.571 | 0.528 | 0.528 | 0.504 | 0.505 |
| 8 | PopulationTransformer (Chau et al. 2024) | - | MIT | 2025-09-25 | 0.545 | 0.689 | 0.677 | 0.576 | 0.628 | 0.509 | 0.519 | 0.509 | 0.523 | 0.519 | 0.513 | 0.505 | 0.509 | 0.508 | 0.499 | 0.492 |
| 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 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.648 | 0.904 | 0.889 | 0.714 | 0.734 | 0.579 | 0.691 | 0.590 | 0.593 | 0.580 | 0.610 | 0.609 | 0.595 | 0.578 | 0.535 | 0.525 |
| 2 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.626 | 0.861 | 0.849 | 0.727 | 0.702 | 0.564 | 0.648 | 0.560 | 0.567 | 0.557 | 0.564 | 0.564 | 0.580 | 0.576 | 0.546 | 0.520 |
| 3 | BrainBERT (frozen; Wang et al. 2023) | - | MIT | 2025-09-25 | 0.581 | 0.743 | 0.631 | 0.572 | 0.692 | 0.509 | 0.661 | 0.571 | 0.580 | 0.572 | 0.556 | 0.559 | 0.527 | 0.534 | 0.506 | 0.497 |
| 4 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.576 | 0.728 | 0.611 | 0.564 | 0.707 | 0.529 | 0.664 | 0.554 | 0.561 | 0.537 | 0.569 | 0.558 | 0.528 | 0.523 | 0.494 | 0.509 |
| 5 | BrainBERT (untrained, frozen) | - | MIT | 2025-09-25 | 0.574 | 0.724 | 0.603 | 0.560 | 0.680 | 0.508 | 0.664 | 0.564 | 0.578 | 0.573 | 0.553 | 0.561 | 0.521 | 0.529 | 0.500 | 0.498 |
| 6 | PopulationTransformer (Chau et al. 2024) | - | MIT | 2025-09-25 | 0.566 | 0.774 | 0.716 | 0.574 | 0.646 | 0.510 | 0.559 | 0.531 | 0.556 | 0.524 | 0.502 | 0.523 | 0.529 | 0.528 | 0.504 | 0.512 |
| 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 | Linear (Laplacian re-referencing + spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.539 | 0.673 | 0.642 | 0.527 | 0.568 | 0.505 | 0.571 | 0.515 | 0.508 | 0.521 | 0.508 | 0.508 | 0.515 | 0.513 | 0.499 | 0.508 |
| 2 | Linear (spectrogram) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.528 | 0.621 | 0.585 | 0.530 | 0.555 | 0.505 | 0.552 | 0.508 | 0.510 | 0.511 | 0.509 | 0.509 | 0.508 | 0.506 | 0.514 | 0.496 |
| 3 | BrainBERT (untrained, frozen) | - | MIT | 2025-09-25 | 0.527 | 0.585 | 0.537 | 0.524 | 0.590 | 0.505 | 0.574 | 0.513 | 0.522 | 0.530 | 0.517 | 0.509 | 0.503 | 0.501 | 0.502 | 0.500 |
| 4 | PopulationTransformer (Chau et al. 2024) | - | MIT | 2025-09-25 | 0.526 | 0.638 | 0.594 | 0.526 | 0.573 | 0.509 | 0.503 | 0.519 | 0.522 | 0.509 | 0.498 | 0.498 | 0.503 | 0.500 | 0.502 | 0.494 |
| 5 | BrainBERT (frozen; Wang et al. 2023) | - | MIT | 2025-09-25 | 0.522 | 0.582 | 0.537 | 0.521 | 0.574 | 0.507 | 0.549 | 0.510 | 0.511 | 0.524 | 0.509 | 0.504 | 0.501 | 0.498 | 0.506 | 0.501 |
| 6 | Linear (raw voltage) | Andrii Zahorodnii | MIT | 2025-09-25 | 0.510 | 0.539 | 0.508 | 0.513 | 0.533 | 0.503 | 0.539 | 0.511 | 0.510 | 0.504 | 0.495 | 0.502 | 0.500 | 0.500 | 0.493 | 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},
}