The sequence-in/sequence-out cerebellar machinery is considered from t
he computational point of view. We outline a learning framework which
discriminates short-term from long-term learning and is able to explai
n single-trial adaptation to unexpected loads.