On the Relation between Sensitivity and Accuracy in In-context Learning
This addresses the unreliability of ICL for real-world applications by improving robustness, though it is incremental as it builds on existing sensitivity analysis.
The paper tackles the problem of in-context learning (ICL) being oversensitive to prompts, which reduces reliability, by finding that prior work underestimated sensitivity due to label bias and showing a negative correlation between sensitivity and accuracy. It proposes SenSel, a selective prediction method that abstains from sensitive predictions, outperforming baselines on ten datasets.
In-context learning (ICL) suffers from oversensitivity to the prompt, making it unreliable in real-world scenarios. We study the sensitivity of ICL with respect to multiple perturbation types. First, we find that label bias obscures the true sensitivity, and therefore prior work may have significantly underestimated ICL sensitivity. Second, we observe a strong negative correlation between ICL sensitivity and accuracy: predictions sensitive to perturbations are less likely to be correct. Motivated by these findings, we propose \textsc{SenSel}, a few-shot selective prediction method that abstains from sensitive predictions. Experiments on ten classification datasets show that \textsc{SenSel} consistently outperforms two commonly used confidence-based and entropy-based baselines on abstention decisions.