First-order ANIL learns linear representations despite misspecified latent dimension

Open in new window