Thursday, January 06, 2005

We sleep because we want to assimilate bimodal priors

Pierre sent me a pointer to a Time article on why we sleep where they seem to talk about the latest research showing that it is not because of your body but rather your brain that you sleep.

In the PBS/Nova story on the secret life of the Brain, they report on experiments done on rats and one experiment where people were reading letters:

Other experiments supplied more direct evidence that sleep is crucial for learning. Human subjects were trained to identify letters that appeared for a blink of an eye on a computer screen. Then, half of the subjects were sent home to sleep, while the other half were deprived of sleep for the entire night, and only then went home to rest. Two days later when all the subjects were already rested and refreshed, the scientists checked their ability to read the flashing letters. None of the participants were tired, and yet the people who went to sleep right after the training performed much better than the ones who went to sleep a day later. This suggests that the night sleep immediately after the activity was crucial for gaining the most from the training session. Without it, the training was much less effective.


More importantly, in this study on learning and reducing behavior to some Bayesian process, the experiment reveals that sleep actual helps remember Bayesian unimodal and bimodal priors. An interesting question I have about this study is: at what point people become experts, when they have learned bimodal priors, tri-modal priors ? is the time needed to learn a multi-modal prior (say n), an O(n^p) process where p > 1 ? Does sleep help reduce the complexity from a high p to a low p ?

While we are it, it looks like that a shift from sleeping late to rising early may signal end of adolescence, many people I know in their forties, are still teen-agers...

No comments:

Printfriendly