Meeting 25: Linear Models

Reading: AIAMA 18.6

Agents often decide on actions depending on a cause in the environmental model. The Naive Bayesian Network, where an effect has several potential causes, forms the basis of classification in a linear hypothesis space. Viewing this as a linear regression is instructive as it forms the basis of mode complex learning methods.

Questions you should be able to answer after reading are:

1. How does linear regression relate to agent decisions?
2. How is optmization used to find optimal hypothesis using training data?
3. Why should sparse models be considered better?
4. What does it mean for a trainign set to be linearly seperable?
5. What is a perceptron?
6. What is the perceptron learning rule?
7. How does one extend perceptrons to non-binary target functions?

Links

• Slides