tribev2-subcortical

TRIBE v2 Subcortical Head (Lahner)

This model provides subcortical fMRI prediction weights for the TRIBE v2 architecture.
It predicts BOLD activity in subcortical brain regions (e.g. hippocampus, amygdala, pallidum) from multimodal inputs (video, audio, text).

---

Model Details

• Developed by: Logan Fernandez
• Model type: Multimodal fMRI encoding model (regression)
• Base model: TRIBE v2
• License: CC BY-NC 4.0

---

Uses

• Predict deep brain activity from naturalistic stimuli
• Study stimulus → brain response relationships
• Analyze emotional response to stimuli
• Make the brain watch reels (See Github repo)

---

Limitations

• Subject-specific behavior for Lahner-style subjects, tuned to short-form media
• Correlation, not causation

---

Training

• Dataset: Lahner2024Bold (BOLD Moments)
• Features: video (V-JEPA), audio (Wac2Vert), text (Qwen3B)
• Projection: subcortical mask

---

Evaluation

• Metric: Pearson correlation
• Result: r = 0.165 (test split)

This reflects the difficulty of predicting subcortical activity, which is noisier and lower resolution than cortical signals.

---

Repository

https://github.com/Enzyme0/homunculus

---

Contact

Logan Fernandez [email protected]

Acknowledgements

This model builds on TRIBE-V2 and is intended as a downstream or derived component built from that work.

Citation

Please cite both this model and TRIBE-V2 when relevant:

@article{dAscoli2026TribeV2,
  title={A foundation model of vision, audition, and language for in-silico neuroscience},
  author={d'Ascoli, St{\'e}phane and Rapin, J{\'e}r{\'e}my and Benchetrit, Yohann and Brookes, Teon and Begany, Katelyn and Raugel, Jos{\'e}phine and Banville, Hubert and King, Jean-R{\'e}mi},
  year={2026}
}