CASE is a top-m linear bandit algorithm with challenger-arm sampling that selects exemplar subsets for in-context learning using up to 7x fewer LLM calls than prior methods.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Sample Efficient Demonstration Selection for In-Context Learning
CASE is a top-m linear bandit algorithm with challenger-arm sampling that selects exemplar subsets for in-context learning using up to 7x fewer LLM calls than prior methods.