Interactive demo · Foundations of Deep Learning · Chapter 1

Polynomial curve fitting: order, data size and weight decay

The ten blue points are the book's own training set (its targets are recovered from the M = 9 column of Table 1.1, see the notes). Fit a polynomial of order M by minimising the sum-of-squares error (1.2), or the regularised error (1.4) with a penalty (λ/2)‖w‖2. Watch the curve, the coefficients, and the root-mean-square error (1.3) on the training set and on a test set of 100 fresh points: a flexible model with no penalty fits the noise.

Training data

Order M

Weight decay ln λ

Meaning of λ

Right-hand plot

Scoreboard

Try these

In the lecture notes