I used the IBL AI Agent (Claude Code) to analyze the BWM dataset and found that
feedback signals are both more widespread and more bidirectional than stimulus signals
across the brain.
Key findings (confirmatory, held-out subjects, n=73 mice, 198 Beryl regions):
- Feedback suppresses 13% of neurons per region vs. 2% for stimulus (p=2.5×10⁻³², ***)
- Stimulus excites more neurons per region than feedback (p=2.7×10⁻⁷, ***)
- Feedback modulates more neurons overall (61.6% vs. 56.9%, p=9.0×10⁻⁵, ***)
- Dorsal raphe shows the largest feedback-over-stimulus excess (+62%)
Full report: https://behkhah.github.io/ibl-ai-agent-reports/feedback-vs-stimulus-breadth/
Analysis used bwm_ephys v1.1.0, stored event_response_features, 50/50 subject-level
exploration/confirmation split. Feedback trials are pooled (reward + error).
I used the IBL AI Agent (Claude Code) to analyze the BWM dataset and found that
feedback signals are both more widespread and more bidirectional than stimulus signals
across the brain.
Key findings (confirmatory, held-out subjects, n=73 mice, 198 Beryl regions):
Full report: https://behkhah.github.io/ibl-ai-agent-reports/feedback-vs-stimulus-breadth/
Analysis used bwm_ephys v1.1.0, stored event_response_features, 50/50 subject-level
exploration/confirmation split. Feedback trials are pooled (reward + error).