#autoencoder

Question 1: Select all that apply

An encoder $f^\ast$ and decoder $g^\ast$ reconstruct every image of a finite training set of distinct images of coloured shapes perfectly: $g^\ast(f^\ast(x_i)) = x_i$ . Nothing else is known about them. Which statements are guaranteed? Select all that apply.

The index $i$ of a training image can be recovered from its code $f^\ast(x_i)$ 1
The colour of a training image can be recovered from its code $f^\ast(x_i)$ 2
Two red training images have closer codes than a red and a blue training image3
The colour of a new image $x$ from the same distribution can be recovered from $f^\ast(x)$ 4

Timed test