Activity 19: Autoencoder exploration#

Chat with folks around you on the following questions:
Question 1. Taking a look at the MNIST examples above, which 2-3 digits do you believe are most “unique”? How would you describe the distinguishing features of these digits?
Your unique digits: TODO
Question 2. Name a pair of digits that you think are most similar. What features are “shared” between these digits?
Your pair: TODO
Next, use the visualization at this link to explore the 2D bottleneck embedding space of an autoencoder trained on MNIST:
Question 3. Are the digits you chose in Question 1 easily “distinguishable” in the latent space? If not, what digits are close to it?
Your answer: TODO
Question 4. Locate the digit pair you chose in Question 2 in the latent space. What do you notice about the relative positions of these digits?
Your answer: TODO
Question 5. Move your mouse around the space. What digit(s) seem to be the hardest to build a “good” 2D representation for?
Your answer: TODO
Question 6. What does the center of the latent space seem to represent?
Your answer: TODO