Stratified Prediction-Powered Inference for Hybrid Language Model Evaluation
This work addresses the need for more efficient and reliable evaluation of language models, particularly when human-labeled data is limited, by providing a method to reduce uncertainty in performance estimates, though it is incremental as it builds on existing prediction-powered inference techniques.
The paper tackles the problem of improving statistical estimates for language model evaluation by proposing Stratified Prediction-Powered Inference (StratPPI), which uses data stratification to tighten confidence intervals, achieving substantially tighter intervals than unstratified methods in cases where automatic labeling performance varies across data distributions.
Prediction-powered inference (PPI) is a method that improves statistical estimates based on limited human-labeled data. PPI achieves this by combining small amounts of human-labeled data with larger amounts of data labeled by a reasonably accurate -- but potentially biased -- automatic system, in a way that results in tighter confidence intervals for certain parameters of interest (e.g., the mean performance of a language model). In this paper, we propose a method called Stratified Prediction-Powered Inference (StratPPI), in which we show that the basic PPI estimates can be considerably improved by employing simple data stratification strategies. Without making any assumptions on the underlying automatic labeling system or data distribution, we derive an algorithm for computing provably valid confidence intervals for population parameters (such as averages) that is based on stratified sampling. In particular, we show both theoretically and empirically that, with appropriate choices of stratification and sample allocation, our approach can provide substantially tighter confidence intervals than unstratified approaches. Specifically, StratPPI is expected to improve in cases where the performance of the autorater varies across different conditional distributions of the target data.