hanqi-qi
commented
1 year ago Hi,
Could I ask some questions about the paper? I am quite confused about the title and the proposed method "Transformation based Adaptation for Ratio prompt learning". While reading your paper, I am confused by the use of the term "in-context learning" and "prompt learning". Can you explain why the title uses "in-context learning" while the proposed method is called "prompt learning"? Why do you use these two terms in these specific contexts?
Hi, Sorry for the late reply. I am hanqi Yan, one author of this paper. Thanks for your interest in our paper.
"In-context" because we involve several training samples into the prompt and feed to the model without fine-tuning the base-model (LLM) parameters. "Prompt Learning" refers that we train of the parameters in the proposed transformer layer which is inserted after the last layer of LLMs. Noted that our trainable parameters do not belong to the typical prompt part, e.g., the input demonstrations, so it causes a bit of confusion. However, if you consider the proposed transformer layers as a generic soft prompt, which can be adaptively trained along with the discrete prompt (i.e., the demonstration), it is kinda of prompt learning. If I may, I would like to suggest that you refer to the two papers regarding "prompt learning" and "in-context learning". I hope you find them useful.
[1] How Does In-Context Learning Help Prompt Tuning [2]Large Language Models Encode Clinical Knowledge
Hi,
Could I ask some questions about the paper? I am quite confused about the title and the proposed method "Transformation based Adaptation for Ratio prompt learning". While reading your paper, I am confused by the use of the term "in-context learning" and "prompt learning". Can you explain why the title uses "in-context learning" while the proposed method is called "prompt learning"? Why do you use these two terms in these specific contexts?