Original link	http://yamlb.wordpress.com/2006/03/13/bayesian-brain-again/
Date	2006-03-13
Status	publish

While writing next post about the meaning of “bayesian” across communities, I decided to make a post just for psychophysics. I’m not qualified, but it seems that a “bayesian rational” agent is a good model for certain human behaviors.

For instance, it seems that humans are optimally combining multiple sensory cues, “optimality” taken in the sense of “bayes optimal”. O. Bousquet discuss this here. Optimality aside, psycho-physicists are proposing that some human behaviors could be explained by considering a probabilistic model+loss function, and with some subjective priors like “velocities are small”.

But, it’s rather difficult to measure “human belief” functions, for that they uses frequencies in repeated (and biased) experiments, which isn’t really what we want to measure. However finding probabilistic models is a promising and exiting research field, even if it’ll be hard to prove something like “the brain is performing bayesian inference” or “neurons are implementing belief propagation”.

Evolutionary speaking, being rational is an advantage, and when you face uncertainty, probabilistic inference can be your friend. Then it’s reasonable to think that we’re performing approximate inference.

To conclude, for neuroscientists, bayesian brain means performing inference on belief states, with subjective priors acquired during evolution/education.

Comments

author: Pierre
date: 2006-03-16 16:47:40

I’d like to add that, in our daily life, that’s not obvious that we always take “rational” decisions. Especially on high level problems, more complex than sensory integration. There are sevral articles in Kahneman (1982) showing that we violate the rules of probability calculs in important ways.