NF-Gym: an offline reinforcement learning benchmark for EEG neurofeedback protocol selection
Year: 2026
Type: software
Abstract: NF-Gym is a benchmark for offline reinforcement learning in closed-loop EEG neurofeedback. Its latent core is a small tabular MDP solved exactly by backward induction, so every result is reported as regret against an external optimum rather than as a within-simulator comparison. The package implements Factored Return Decomposition (FRD), an unstructured return-decomposition baseline, potential-based reward shaping with the terminal-potential convention required for episodic policy invariance, and the numerical checks for Propositions 1, 3, 4 and 5 of the accompanying paper.
Author: Eroglu, Gunet
Institution: Bahçeşehir University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204147496