Expand description
Multi-layer fusion operator (Section 7, Paper 4).
Implements deep Bayesian fusion as a multi-layer network:
l^(k) = g( l^(k-1) + sum_j logit(P_j^(k)) )
P_final = sigmoid(l^(K))The internal channel map keys per-document feature vectors. Each
Layer variant updates that map, including signal, dense, convolutional,
recurrent, normalization, attention, and graph-aware propagation / pooling
layers.