2024-07-11 –, For Loop (3.2)
SequentialSamplingModels.jl is a package for simulating and evaluating a large class of human decision-making models called sequential sampling models (SSMs). We will briefly introduce SSMs and their utilization throughout cognitive science, illustrate some of the main features of the current package version, and discuss future developments.
SequentialSamplingModels.jl provides a unified interface for simulating and evaluating sequential sampling models (SSMs) in Julia.
SSMs describe human decision making as a stochastic and dynamic evidence accumulation process. The models assume that evidence accumulates over time, starting from a certain position, until it crosses a boundary, triggering a corresponding response (Ratcliff & McKoon, 2008; Ratcliff, 1978). SSMs have seen wide-ranging applications in cognitive psychology & neuroscience due to the interpretability of parameters as representative of multiple components of information processing and their ability to form a generative model for reaction time distributions for responses.
SequentialSamplingModels.jl has been integrated with the probabilistic programming ecosystem through:
• Distributions.jl: functions for describing probability distributions
• Plots.jl: extended plotting tools for SSMs
• Turing.jl: Bayesian parameter estimation
In our talk, we will briefly introduce the theory and applications of SSMs and showcase some of the main features of the current version of the package through a quick tutorial including how to:
• install the package
• define an SSM
• simulate data from the constructed SSM
• conduct parameter estimation with Turing
Lastly, with many SSMs having likelihoods that cannot be computed efficiently, simulation-based inference has recently been proposed as a solution. We highlight areas for future development, aiming to incorporate likelihood approximation networks (which use neural networks to predict log-likelihoods from data and parameters) using packages like Flux.jl, and invite suggestions for other machine learning-based approaches to parameter estimation.
For a more comprehensive exploration of the package functionalities and further details on future developments, the user is invited to consult the package README, documentation, and issues.
Kiante Fernandez is pursuing a PhD in Computational Cognition at the University of California, Los Angeles, under the supervision of Dr. Ian Krajbich. His primary research interest is in modeling human decision-making in economic settings.