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

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

rewardGym - a framework for streamlining experiments in cognitive neuroscience

Simon R Steinkamp¹, David Meder², Oliver James Hulme³; ¹Copenhagen University Hospital - Amager and Hvidovre, Copenhagen, Denmark , ²Copenhagen University Hospital - Amager and Hvidovre, ³Copenhagen University

Presenter: Simon R Steinkamp

Despite significant efforts, psychological and cognitive sciences currently lack standardized frameworks for setting up tasks and models in ways that can be integrated with data, generalized over tasks, and shared between researchers. Typically, researchers program an experimental task, assess how a theoretical model would solve it, and analyze behavioral data, all using custom code. Due to a lack of standardization, cognitive models are often experiment-specific, making it difficult to test the generalizability of models across different tasks and increasing the burden of testing existing models on new tasks. Here, we present a standardized framework that addresses these challenges. We build on the Gymnasium standard from reinforcement learning (RL), which defines how artificial agents interact with computational environments. This standard helps us establish a common graphical language for different tasks that captures the connections between states, actions, and rewards. This representation is further inspired by neuro-nav, which extends the Gymnasium framework to classical neuroscience experiments and focuses on neurally plausible RL. By expressing tasks in a formal language, it is possible to build libraries of models where agents can perform each task. This allows standardized software to perform parameter inference and model comparison on real and synthetic data. What distinguishes our framework is its focus on running experiments. We provide a high degree of control over the environments (e.g., stimulus order) and a direct way to augment the graphical representations with stimulus information. This allows for a direct transition from simulations with artificial agents to running experiments with human participants using PsychoPy. Additionally, we provide logging utilities that save data in BIDS format, which is standard in neuroimaging. With this framework, we hope to make psychological and cognitive science more reproducible and robust.

Topic Area: Reward, Value & Social Decision Making

Extended Abstract: Full Text PDF