Up to section 5: Loss / Accuracy on TEST data lower / higher than on TRAIN data ??

[aiqei9]

I've been following this course reasonably well up to CNNs but there's something bugging me: up to now, in pretty much all train & test exercises, the Test Loss seems lower throughout each epoch than the Train Loss, while the reverse happens with respect to Accuracy, all of which is exactly opposite what one should expect.

Any ideas? Am I interpreting something incorrectly? Thanks!