Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Reinforcement Learning via Brain Feedback (RLBF)

Traditional fMRI uses fixed paradigms. RLBF reverses the usual direction of inference: neural responses guide exploration of a stimulus space via reinforcement learning, so that stimuli can be adapted in real time to optimise a chosen brain target (regional activity or a multivariate signature).

Reinforcement Learning via Brain Feedback closed-loop schematic

An AI paradigm generator presents a task/stimulus; real-time fMRI yields a neural response that serves as reward feedback for the RL agent, which updates the next stimulus — closing the loop.

Key publication

Gallitto et al. — Reinforcement Learning via Brain Feedback for real-time fMRI-based adaptive stimulus generation

The accompanying Python framework integrates real-time fMRI processing, RL agents, adaptive stimulus generation, simulation testing, and experiment monitoring. In a proof-of-concept (N=10), RLBF adapted checkerboard contrast and frequency within a single ~10-minute session to maximise V1 responses.

Software

References
  1. Gallitto, G., Englert, R., Kincses, B., Kotikalapudi, R., Li, J., Hoffschlag, K., Ali, S., Bingel, U., & Spisak, T. (2026). Reinforcement Learning via Brain Feedback for real-time fMRI-based adaptive stimulus generation. 10.64898/2026.08.08.743648