• Speaker: Brandon Turner, Stanford University
  • Title: ABCDE: A practical likelihood-free Bayesian analysis technique with applications to mathematical models of cognition
  • Location: Room V206, Mathematics Building (Callaghan Campus) The University of Newcastle
  • Time and Date: 4:00 pm, Wed, 24th Oct 2012
  • Abstract:

    Many cognitive models derive their predictions through simulation. This means that it is difficult or impossible to write down a probability distribution or likelihood that characterizes the random behavior of the data as a function of the model's parameters. In turn, the lack of a likelihood means that standard Bayesian analyses of such models are impossible. In this presentation we demonstrate a procedure called approximate Bayesian computation (ABC), a method for Bayesian analysis that circumvents the evaluation of the likelihood. Although they have shown great promise for likelihood-free inference, current ABC methods suffer from two problems that have largely revented their mainstream adoption: long computation time and an inability to scale beyond models with few parameters. We introduce a new ABC algorithm, called ABCDE, that includes differential evolution as a computationally efficient genetic algorithm for proposal generation. ABCDE is able to obtain accurate posterior estimates an order of magnitude faster than a popular rejection-based method and scale to high-dimensional parameter spaces that have proven difficult for the current rejection-based ABC methods. To illustrate its utility we apply ABCDE to several well-established simulation-based models of memory and decision-making that have never been fit in a Bayesian framework.

    AUTHORS: Brandon M. Turner (Stanford University) Per B. Sederberg (The Ohio State University)

  • [Permanent link]