Implicit Behavioral Decoding from Next-Step Spike Forecasts at Population Scale
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911679376261120 |
|---|---|
| author | Minnick, John R. Gonzalez-Ferrer, Jesus Hussain, Kamran Geng, Jinghui Robbins, Ash Mostajo-Radji, Mohammed A. Haussler, David Eshraghian, Jason Teodorescu, Mircea |
| author_facet | Minnick, John R. Gonzalez-Ferrer, Jesus Hussain, Kamran Geng, Jinghui Robbins, Ash Mostajo-Radji, Mohammed A. Haussler, David Eshraghian, Jason Teodorescu, Mircea |
| contents | Closed-loop brain-computer interfaces often require both a forecast of upcoming neural population activity and a readout of the animal's behavioral state. A single Mamba forecaster, trained only on next-step spike counts at Neuropixels scale, can deliver both in one forward pass. A lightweight per-session linear head reading the model's predicted rates decodes behavior better than the same linear classifier reading the raw spike counts, under matched temporal context. We test on the Steinmetz visual-discrimination benchmark, which spans 39 sessions, roughly 27,000 neurons, and 1,994 held-out trials. Across three training seeds, Mamba's predicted rates decode mouse choice at 75.7$\pm$0.2% trial vote, roughly 2.3 times chance level, and stimulus side at 66.1$\pm$0.6%, about twice chance. Compared to a matched 500 ms-context linear decoder on the raw spike counts, Mamba wins at trial vote by 4-6 pp on response and 4-6 pp on stimulus side. A session-start calibration block of about 100-150 trials brings the readout within 1-2 pp of asymptote, and the full pipeline fits inside the 50 ms bin budget on workstation-class GPUs typical of tethered chronic Neuropixels recordings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_12999 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Implicit Behavioral Decoding from Next-Step Spike Forecasts at Population Scale Minnick, John R. Gonzalez-Ferrer, Jesus Hussain, Kamran Geng, Jinghui Robbins, Ash Mostajo-Radji, Mohammed A. Haussler, David Eshraghian, Jason Teodorescu, Mircea Neurons and Cognition Machine Learning Closed-loop brain-computer interfaces often require both a forecast of upcoming neural population activity and a readout of the animal's behavioral state. A single Mamba forecaster, trained only on next-step spike counts at Neuropixels scale, can deliver both in one forward pass. A lightweight per-session linear head reading the model's predicted rates decodes behavior better than the same linear classifier reading the raw spike counts, under matched temporal context. We test on the Steinmetz visual-discrimination benchmark, which spans 39 sessions, roughly 27,000 neurons, and 1,994 held-out trials. Across three training seeds, Mamba's predicted rates decode mouse choice at 75.7$\pm$0.2% trial vote, roughly 2.3 times chance level, and stimulus side at 66.1$\pm$0.6%, about twice chance. Compared to a matched 500 ms-context linear decoder on the raw spike counts, Mamba wins at trial vote by 4-6 pp on response and 4-6 pp on stimulus side. A session-start calibration block of about 100-150 trials brings the readout within 1-2 pp of asymptote, and the full pipeline fits inside the 50 ms bin budget on workstation-class GPUs typical of tethered chronic Neuropixels recordings. |
| title | Implicit Behavioral Decoding from Next-Step Spike Forecasts at Population Scale |
| topic | Neurons and Cognition Machine Learning |
| url | https://arxiv.org/abs/2605.12999 |