Product-of-Experts Training Reduces Dataset Artifacts in Natural Language Inference

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Mathew, Aby Mammen
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911611296415744
author Mathew, Aby Mammen
author_facet Mathew, Aby Mammen
contents Neural NLI models overfit dataset artifacts instead of truly reasoning. A hypothesis-only model gets 57.7% in SNLI, showing strong spurious correlations, and 38.6% of the baseline errors are the result of these artifacts. We propose Product-of-Experts (PoE) training, which downweights examples where biased models are overconfident. PoE nearly preserves accuracy (89.10% vs. 89.30%) while cutting bias reliance by 4.71% (bias agreement 49.85% to 45%). An ablation finds lambda = 1.5 that best balances debiasing and accuracy. Behavioral tests still reveal issues with negation and numerical reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2604_19069
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Product-of-Experts Training Reduces Dataset Artifacts in Natural Language Inference
Mathew, Aby Mammen
Computation and Language
Artificial Intelligence
68T50, 68T07
I.2.7; I.2.6
Neural NLI models overfit dataset artifacts instead of truly reasoning. A hypothesis-only model gets 57.7% in SNLI, showing strong spurious correlations, and 38.6% of the baseline errors are the result of these artifacts. We propose Product-of-Experts (PoE) training, which downweights examples where biased models are overconfident. PoE nearly preserves accuracy (89.10% vs. 89.30%) while cutting bias reliance by 4.71% (bias agreement 49.85% to 45%). An ablation finds lambda = 1.5 that best balances debiasing and accuracy. Behavioral tests still reveal issues with negation and numerical reasoning.
title Product-of-Experts Training Reduces Dataset Artifacts in Natural Language Inference
topic Computation and Language
Artificial Intelligence
68T50, 68T07
I.2.7; I.2.6
url https://arxiv.org/abs/2604.19069