[Adapt] [Seminar] Demonstration Selection for In-Context Learning

[贾琪]

Hi Adapters,The striking language skills and world knowledge embedded in large pre-trained language models have recently led to in-context learning, where the model learns to make a prediction for a test input by conditioning on a handful of task-specific examples without any parameter updates. However, performance has been shown to strongly depend on the selected training examples (demonstrations). Discovering the best demonstration combination proves to be a formidable challenge, given the existence of $\sum_{k=1}^{N}\mathrm{C}_{N}^{k}k!$ distinct candidates. This week, I'll introduce a group of methods that improve the performance of ICL by demonstration (example) selection for approximating the most suitable demonstrations.Hope you find this talk interesting and useful.Best Regards,
Angel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20230329/e0addb1e/attachment.htm>