Boosting Vision-Language Models with Transduction | Read Paper on Bytez