Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Towards generative AI-based fMRI paradigms: reinforcement learning via real-time brain feedback

Giuseppe Gallitto¹, Robert Englert, Balint Kincses, Raviteja Kotikalapudi, Jialin Li¹, Kevin Hoffschlag, Sulin Ali, Ulrike Bingel, Tamas Spisak¹; ¹Universität Duisburg-Essen

Presenter: Giuseppe Gallitto

Traditional fMRI’s reliance on fixed task paradigms perpetuates the reverse inference problem, limiting specificity in brain-behavior mapping. We present Reinforcement Learning via Brain Feedback (RLBF), an approach reversing the direction of inference by using real-time fMRI and reinforcement learning to dynamically adjust stimuli – optimizing the ‘stimulus space’. In a visual cortex proof-of-concept study (N=10), the algorithm successfully optimized checkerboard parameters within 35 trials, improving brain prediction from chance-level to a mean absolute percentage error (MAPE) of 12.7% (SD:6.3%; inter-trial improvement of 0.6%) and achieving stable convergence, despite fMRI noise. Stimulation optimization revealed a preference for maximum contrast stimuli at 18Hz (+/-10Hz across participants) – aligning with known visual processing properties – and demonstrated the potential for brain activity to effectively tune AI models, offering new avenues for personalized experimental design and rigorous testing of reverse inference claims.

Topic Area: Reward, Value & Social Decision Making

Extended Abstract: Full Text PDF