“…In recent years, the development of LMs has revolutionized the field of NLP [1,2], leading to remarkable progress in a range of downstream tasks, including text classification [3,4], information retrieval [5,6,7,8], and multi-modalities [9,10,11]. While LMs have shown impressive performance in many tasks, there has been a debate over the effectiveness of simple data augmentation (DA) techniques, such as back-translation, for improving the FT performance.…”