PLMs face challenges in few-shot scenarios due to over-multitudinous conceptual knowledge. BayesPrompt addresses this by approximating debiased factual distributions for downstream domains.
