Slow Learning Rates


Learning is something that we do all the time. This isn’t the focused effort that we might put into a course, but rather the everyday problems of the environment that we’re working in. This often involves us in learning complex systems and what response we need to make to them. We do this without realizing that we’re acquiring knowledge. This is termed implicit learning.

In some instances, this type of learning could be a matter of survival. For example if we are learning to fly a plane, responding to turbulent air requires a correct prediction of movement. Of course, there are many other examples that might have less dire consequences of slower learning.

Thoughts about the learning process bring some questions to mind. One of these is what affects our learning rate? Dienes et al set themselves to have a look at this and studied things like mood. Their paper is in Friday’s edition of the Public Library of Science (1).

Being psychologists, they went to the computer screen and pressing buttons for the answers. Basically they had colored squares appearing on the screen and the human lab rats had to press buttons indicating their prediction of which side of the screen the image would appear.

They put in variations and controls in an attempt to isolate the good data. The basic premise was that if the lab rat thought that they could see a pattern, they would be using a series of the previous images to make their decision. This corresponded to a slow learning rate in that the correlation with earlier occurrences would fade slowly with number of tests.

Now a fast learning rate meant that there would be some weight on the previous occurrence but little or none on events prior to that, so the correlation would fall off rapidly with number of trials. The decrease was taken to be exponential with a steeper decay (faster learning rate) for those candidates who only let the previous test affect their choice and not the third and fourth test back in time.

The results of the controls compared to an amnesia patient showed that the person suffering from amnesia had a faster learning rate, that is, he took less notice of previous result on his choice. Perhaps he couldn't remember?

The experimenters introduced another twist in that to some participants they displayed a happy face prior to their choice and to others they displayed a sad face. This influenced the mood of the participants who were being tested. The results showed that the happy people had a longer correlation decay, that is, the learned more slowly.

The big message that I take away here is that should I go gambling on something random like a coin toss, I’d better make sure that I’m unhappy or suffering from amnesia to avoid getting caught up in trying to see patterns in random events and fall foul of the gambler’s fallacy.

  1. http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0033400


Leave a Reply