Simpson's Bias in NLP Training

TL;DR

This paper investigates Simpson's bias in NLP training, revealing that common sample-level loss functions can be inconsistent with true population metrics, leading to sub-optimal model performance.

Contribution

It provides a systematic theoretical and experimental analysis of Simpson's bias in NLP, highlighting its impact on model training and evaluation.

Findings

Popular loss functions may not align with true evaluation metrics.

Models optimized with certain losses can be substantially sub-optimal.

The paper connects Simpson's bias with classical statistical paradoxes.

Abstract

In most machine learning tasks, we evaluate a model $M$ on a given data population $S$ by measuring a population-level metric $F(S;M)$. Examples of such evaluation metric $F$ include precision/recall for (binary) recognition, the F1 score for multi-class classification, and the BLEU metric for language generation. On the other hand, the model $M$ is trained by optimizing a sample-level loss $G(S_t;M)$ at each learning step $t$, where $S_t$ is a subset of $S$ (a.k.a. the mini-batch). Popular choices of $G$ include cross-entropy loss, the Dice loss, and sentence-level BLEU scores. A fundamental assumption behind this paradigm is that the mean value of the sample-level loss $G$, if averaged over all possible samples, should effectively represent the population-level metric $F$ of the task, such as, that $\mathbb{E}[ G(S_t;M) ] \approx F(S;M)$. In this paper, we systematically investigate the above assumption in several NLP tasks. We show, both theoretically and experimentally, that some popular designs of the sample-level loss $G$ may be inconsistent with the true population-level metric $F$ of the task, so that models trained to optimize the former can be substantially sub-optimal to the latter, a phenomenon we call it, Simpson's bias, due to its deep connections with the classic paradox known as Simpson's reversal paradox in statistics and social sciences.