Fig. 1: Model architecture for deep multi-task self-supervised pretext training to obtain RNA contact map candidates.
From: RNA contact prediction by data efficient deep learning
Boxes represent model building blocks, edges information flow, and brackets tensor shapes. L is the length of the sequence, E the evolutionary dimension, i.e., available alignments, N the number of attention encoder blocks, H the number of attention heads, C the number of arrangements of the jigsaw, and D a parametric embedding dimension.
