Questions on the proof. of Lemma 2.

[libo-huang]

Dear Mr. Ye,

I am happy to meet you at this project through your masterpieces, Learning Latent Representations Across Multiple Data Domains Using Lifelong VAEGAN, published in ECCV 2020. But I am confused about the proof of Lemma 2 in the Appendix. Specifically, it seems there is a type error in Eq. (3). Replace $B{1^{K-1}}$ with $B{1^K}$ seems better. If this correct is sound, how can we deduce the following results till Eq. (4)?

Many thanks for your time, and looking forward to your reply.

[dtuzi123]

I agree with you, I will replace $B{1^{K-1}}$ with $B{1^K}$ in Appendix. For Eq.4, we can sum up all task risks (i=1,...,K) by using the end of Eq.3.