C2

Question 1

Matrix factorization with gradient descent

Answer 1

1. initialize all variables at random
2. iterate over all records (user, item, rating):
   - calculate the direction of the steepest descend of function f()
   - make a step of size L_rate in this direction

till convergence or a stop criterion is met

Question 2

regularization

Answer 2

extra mechanism that controls the complexity of the model (lambda): prevent overfitting

eg. punish the polynomial for having too high values of coefficients