The physics of decision making: stochastic differential equations as models for neural dynamics and evidence accumulation in cortical circuits

Philip Holmes, Philip Eckhoff, KongFatt Wong-Lin, Rafal Bogacz, Miriam Zackenhouse, Jonathan Cohen

    Research output: Book/ReportBookpeer-review

    Abstract

    We describe how drift-diffusion (DD) processes – systems familiar in physics – can be used to model evidence accumulation and decision-making in two-alternative, forced choice tasks. We sketch the derivation of these stochastic differential equations from biophysically-detailed models of spiking neurons. DD processes are also continuum limits of the sequential probability ratio test and are therefore optimal in the sense that they deliver decisions of specified accuracy in the shortest possible time. This leaves open the critical balance of accuracy and speed. Using the DD model, we derive a speed-accuracy tradeoff that optimizes reward rate for a simple perceptual decision task, compare human performance with this benchmark, and discuss possible reasons for prevalent sub-optimality, focussing on the question of uncertain estimates of key parameters. We present an alternative theory of robust decisions that allows for uncertainty, and show that its predictions provide better fits to experimental data than a more prevalent account that emphasises a commitment to accuracy. The article illustrates how mathematical models can illuminate the neural basis of cognitive processes.
    Original languageEnglish
    PublisherWorld Scientific Publishing
    Number of pages123
    ISBN (Print)978-981-4304-63-4
    DOIs
    Publication statusPublished (in print/issue) - Mar 2010

    Bibliographical note

    Plenary lecture at the XVIth International Congress on Mathematical Physics, Prague, Czech Republic, Aug 2-8, 2009

    Fingerprint

    Dive into the research topics of 'The physics of decision making: stochastic differential equations as models for neural dynamics and evidence accumulation in cortical circuits'. Together they form a unique fingerprint.

    Cite this