de Groot | CUBE 216

at

Jul 19 @ 10:00 CEST - Jul 19 @ 13:30 CEST

Public session

Reinforcement learning (RL) algorithms have proven to be exceptionally effective in modeling human value-based learning and decision-making behaviors. This workshop offers an in-depth introduction to RL algorithms and their application in modeling human decision-making behaviors. Starting with the fundamentals, participants will learn model-free RL algorithms applied to a multi-armed bandit task. The workshop will then advance to explore two key extensions; (1) hierarchical RL modeling, where a sequence of action is required to complete a goal, and (2) model-based RL modeling where action-values are computed based on mental simulation of possible state transitions. Throughout the workshop, participants will engage in practical, hands-on coding exercises. These will include Bayesian parameter recovery to identify RL agents' parameters using Stan. Additionally, attendees will experience parameter estimation applied to existing human empirical data. The workshop is tailored for a wide audience from those with a basic understanding of programming and statistics to experienced researchers in cognitive modeling seeking to deepen their understanding of basic RL modeling techniques. By the end of the session, participants will have acquired basic theoretical knowledge and practical skills in implementing RL models, setting the stage for further exploration and application of RL in various domains of cognitive science and beyond.

Jul 20 @ 12:00 CEST - Jul 20 @ 13:40 CEST

In-person session

The recent proliferation of intensive longitudinal data allows researchers to investigate how psychological processes evolve at an unprecedented temporal resolution. However, these large time series datasets come with unique challenges that require advanced modeling solutions to enable reliable inferences. In this symposium, we outline several advancements in time series modeling and show how they improve our understanding of psychological processes using empirical applications. The first talk discusses how Dynamic Structural Equation Models capture temporal dynamics and outlines their psychometric properties. The second talk shows how DSEMs can be extended to capture cognitive dynamics across multiple timescales contemporaneously. The third talk illustrates how adequate modeling of night gaps in ESM can improve our understanding of daytime versus nighttime dynamics in psychological processes. The fourth talk advocates a new standard for time series modeling allowing temporal dynamics to differ based on the time series value. The fifth talk combines time series models with Hidden Markov Models to capture mood states. Together, these five talks outline how advanced time series models help improve our understanding of a wide range of psychological processes and provide openly-available modeling code for researchers to apply the models themselves.

Jul 20 @ 14:00 CEST - Jul 20 @ 15:40 CEST

In-person session

Jul 21 @ 10:00 CEST - Jul 21 @ 16:20 CEST

In-person session

Jul 22 @ 10:00 CEST - Jul 22 @ 11:20 CEST

In-person session

Jul 22 @ 11:40 CEST - Jul 22 @ 17:00 CEST

In-person session