Filter likelihoods and exhaustive learning