b) The jointly-learned state-action embedding approach allows for more efficient exploration by leveraging similarities between actions in the embedding space