Topics covered in Recitation 3 (Friday, Sep. 24): 1. Review of two active learning techniques (minimising weight covariance matrix or minimising output variance) discussed in class. 2. Derivation of expressions for (and explanation of) trace and determinant of weight covariance matrix 3. Working through the math for a simple linear regression example to show that minimising the determinant implies selecting input points spaced as far apart from each other as possible. 4. Proof that querying new inputs reduces the output variance for all inputs (using some results for positive semidefinite matrices). 5. Exam question from 2003 midterm: active learning in a classification problem (logistic regression) 6. Clarification of a few points regarding problem set 1.