Skip to main content
. Author manuscript; available in PMC: 2022 Sep 23.
Published in final edited form as: KDD. 2022 Aug 14;2022:4402–4412. doi: 10.1145/3534678.3539163

Figure 3:

Figure 3:

We use an auto-encoder represent the CFP sequences as vectors. Encoder takes visits’ feature xt and time gaps Δt as inputs. Decoder output the previous AMD steps st at different time. The memory vector CT contains the CFP sequence information of the eye. Finally, we group eye representations CT into AMD subphenotypes.