Contributions: Abstract
There is a close link between optimal experimental design and certain models for "directed attention" in cognitive science. The concept of information conveniently brings the two areas together. Put in a deceptively elementary way: to learn well one needs to direct attention in a way to optimally increase information. The mathematical formulation is critically dependent on one's model for observation. In active mode, which is similar to controlled experimentation, one has the right to change the conditional environment radically. At the other extreme one accepts data passively. But even in the latter case under a "computational model" for learning one might only accept data which is "directed". If observation is cheap, why accept data with only a small informational content if computation is expensive? But being too selective interferes with the ability to learn the background. Information theoretic models suggest solutions to these problems.
|