confused about 2020ICML-weakly supervised disentanglement

[junkangwu]

Hi, weakly-supervised disentanglement is a nice work. But I am still confused about some concept. Looking forward to your explanation.

• in the paper at page 5, you compare the work with prior work. "Our approach critically differs in the sense that S is not known and needs to be estimated for every pair of images." May I ask you about the impletation of S estimation? It seems that I have ignore some crucial details.
• What is the meaning of Rnd? Is it an abbreviations?

[schneimo]

Hi, even though I am not an author I think I can still answer your questions:

• You can find the answers to this question on page 4: "To obtain an estimate of S we therefore choose for every pair (x_1, x_2) the d−k coordinates with the smallest D_KL"
• I think Rnd is simply the abbreviation for Random

[junkangwu]

Hi, even though I am not an author I think I can still answer your questions:

• You can find the answers to this question on page 4: "To obtain an estimate of S we therefore choose for every pair (x_1, x_2) the d−k coordinates with the smallest D_KL"
• I think Rnd is simply the abbreviation for Random

Hi, Thanks a lot！ I am unfamiliar to the area of the disentanglement at the last time. I am sorry to bother you again！ In the training of weakly-vae, pairs are adopted to discover the shared dimensions and specific dimensions. However, if I construct the similar pairs to train, it is necessary to keep them at the valid or test procedure? In other words, What should I do to generate the output focus on the single instance in validate dataset and test dataset? In fact, most of the test instances always are not observed before test procedure.

Best！