Stronger Random Baselines for In-Context Learning
This work addresses evaluation challenges for researchers in NLP by providing a more rigorous baseline to prevent inflated claims, though it is incremental as it builds on existing baseline methods.
The paper tackles the problem of overestimating in-context learning performance due to weak random baselines in evaluation, and shows that a stronger baseline based on expected maximum accuracy across random classifiers reveals over 20% of few-shot results as not significant when applied to 16 tasks and six models.
Evaluating the in-context learning classification performance of language models poses challenges due to small dataset sizes, extensive prompt-selection using the validation set, and intentionally difficult tasks that lead to near-random performance. The standard random baseline--the expected accuracy of guessing labels uniformly at random--is stable when the evaluation set is used only once or when the dataset is large. We account for the common practice of validation set reuse and existing small datasets with a stronger random baseline: the expected maximum accuracy across multiple random classifiers. When choosing the best prompt demonstrations across six quantized language models applied to 16 BIG-bench Lite tasks, more than 20% of the few-shot results that exceed the standard baseline do not exceed this stronger random baseline. When held-out test sets are available, this stronger baseline is also a better predictor of held-out performance than the standard baseline, avoiding unnecessary test set evaluations. This maximum random baseline provides an easily calculated drop-in replacement for the standard baseline.