Bayesian use of historical knowledge in perceptual trajectory extrapolation:
Computational and neural mechanisms
-
1
University of Oxford, FMRIB Centre, United Kingdom
To interact effectively with the environment, brains need to extract the predictive value of information incident upon their senses. Prediction can involve both computational processes associated with motor control, and with reward-based reinforcement learning.
In motor control, forward models are used to extrapolate the results of motor commands. Forward models may be computed by the cerebellum [1], and their output is evident in parietal cortex [2]. In other situations, (e.g. foraging) statistical properties of the environment must be extracted, using reinforcement learning. The striatum is important for learning, and the resulting statistical maps are evident in parietal cortex [3]. Optimal performance in a noisy environment may be achieved by Bayesian combination of forward-modelling and a historical prior acquired through reinforcement learning [4].
We used fMRI to investigate the neural mechanisms by which these different computations are processed and combined. Participants predicted the endpoints of visually observed trajectories - a forward modelling/extrapolation task. Probabilistic/historical information was also informative as the (Gaussian) distribution of trajectory-endpoints could be learnt over many trials. The relative value of information sources was manipulated trial-to-trial by adding noise to the observed trajectory and/or changing the variance of the prior/historical distribution. We modelled participants’ changing beliefs about the historical prior using a Bayesian computer learner.
Behaviourally, participants’ endpoint predictions were better modelled by the Bayesian posterior than the prior or forward-model alone. Initial analyses of fMRI data indicate differential involvement of motor systems (including cerebellum, IPS and FEF) when forward-modelling is the more reliable computation.
References
1. Miall et al 2007 PLoS Biol. 2007 5(11):e316.
2. Mulliken et al 2008, PNAS 105: 8170-7.
3. Yang and Shadlen 2008, Nature 447:1075-80.
4. Kording and Wolpert 2004, Nature 427 :244-7.
Conference:
Computations, Decisions and Movement, Giessen, Germany, 19 May - 22 May, 2010.
Presentation Type:
Poster Presentation
Topic:
Posters
Citation:
O'Reilly
J and
Behrens
T
(2010). Bayesian use of historical knowledge in perceptual trajectory extrapolation:
Computational and neural mechanisms.
Front. Comput. Neurosci.
Conference Abstract:
Computations, Decisions and Movement.
doi: 10.3389/conf.fnins.2010.01.00012
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
01 Feb 2010;
Published Online:
01 Feb 2010.
*
Correspondence:
Jill O'Reilly, University of Oxford, FMRIB Centre, Oxford, United Kingdom, joreilly@fmrib.ox.ac.uk