University of Cambridge > > Computational and Biological Learning Seminar Series > Seeing Patterns in Randomness: Irrational Superstition or Adaptive Behavior?

Seeing Patterns in Randomness: Irrational Superstition or Adaptive Behavior?

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Zoubin Ghahramani.

Just as visual illusions reveal principles and mechanisms underlying natural visual processing, “cognitive illusions” can provide similar insight into the computations underlying decision-making. In a variety of behavioral tasks involving randomized sequences of stimuli, subjects exhibit an odd “sequential effect”: their responses are facilitated by chance runs of repetitions and alternations in stimulus identity, and are hampered by violations of such transient local patterns. Such behavior is sub-optimal in the context of the task, since these transient patterns are spurious and have no true predictive power of future stimuli. We hypothesize that this apparently superstitious behavior may in fact reflect the engagement of adaptive mechanisms that are critical for responding to environments containing statistics that change over time. We use a normative Bayesian framework to show that a prior belief in changing statistics is sufficient to produce the type of sequential effects observed in subjects’ behavior. The Bayesian algorithm is shown to have an approximately linear dependence on the history of past observations, where the dependence decays exponentially into the past; this exponential decay is also apparent in human and monkey behavior, as we found by re-analyzing data from published studies. We show that the necessary computations can be implemented by neurons with dynamics previously thought to underlie a variety of sensory processing and decision making computations. Our work provides a unifying account of how decision-making SHOULD adapt to a changing environment, a functional theory for WHY subjects exponentially discount the past, as well as a principled hypothesis of HOW neurons can implement the necessary computations.

This talk is part of the Computational and Biological Learning Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2023, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity