[cs231n_review#Lecture 2-4] Cross-Validation

 

idea # 4 Cross-validation

Where your algorithm is able to see the labels of the training set, but for the validation set, your algorithm doesn't have direct access to the labels. We only use the labels of the validation set to check how well our algorithm is doing.

(Cross-Validation can alleviates over-fitting problem)

these things like Euclidean distance, or L1 distance, are really not a very good way to measure distances between images. These, sort of, vectorial distance functions do not correspond very well to perceptual similarity between images.

 

we've actually shifted down by a couple pixels, or tinted the entire image blue. And, actually, if you compute the Euclidean distance between the original and the boxed, the original and the shuffled, and original in the tinted, they all have the same L2 distance. Which is, maybe, not so good because it sort of gives you the sense that the L2 distance is really not doing a very good job at capturing these perceptional distances between images.