Tag
This paper presents gradCSCG, a differentiable reformulation of the Clone-Structured Causal Graph algorithm for end-to-end learning of cognitive maps directly from image sequences, combining it with a vector-quantized variational autoencoder.