Mapping fNIRS to Agent Performance
Towards reinforcement learning from neural feedback
This project investigates the relationship between passive neural signals (fNIRS) and artificial agent performance. We ask whether implicit hemodynamic responses can identify how well an agent is performing and act as feedback for a human-in-the-loop reinforcement learning agent.
We analyze correlations between hemodynamic response in the prefrontal cortex (PFC) and agent behavior. If a user’s neural activity can track whether an agent is performing well, we can use those signals to support human-in-the-loop reinforcement learning without explicit demonstration or guidance.
The work connects cognitive neuroscience sensing (fNIRS) with RL from human feedback, on the idea that RL policies may be shaped by implicit fNIRS data.