Unfortunately(?), unlike babies, neural networks don't get bored, and so we can't apply looking-time-based measures. What do we do? We extract embeddings (vector representations) of the stimuli and test the similarity between embeddings (see NN-infant comparison below, 3/N).
