When Sample Selection Bias Precipitates Model Collapse

2026-06-11

AI Summary

This work addresses the challenge of model collapse and rapidly decaying generation diversity—following a power-law decay—under low-resource and data-isolated settings, where sample selection based on local biased reference distributions exacerbates instability. The study is the first to uncover this degradation mechanism and proposes a collaborative proxy reference construction method that requires no sharing of raw data. By aligning distributions across isolated data silos via Wasserstein distance, the approach constructs a globally consistent reference signal to guide recursive training of synthetic data. Theoretical analysis and distributed experiments demonstrate that, compared to local reference strategies, the proposed method effectively mitigates diversity degradation and substantially enhances model stability and generalization capability.

Abstract

The proliferation of recursive training on synthetic data can alleviate data scarcity but risks model collapse, where repeated training erodes distributional tails and homogenizes outputs. Data selection is widely viewed as a remedy, yet its reliability depends critically on the reference distribution used by the verifier. We show that in low-resource verification regimes, where each verifier observes only a small, fragmented, and biased slice of the target manifold, selection itself becomes biased. This situation naturally arises in low-resource data silos such as healthcare consortia or proprietary financial institutions, where raw data cannot be pooled and local references are inherently incomplete. As a result, selection preferentially retains samples aligned with the local manifold while pruning globally relevant tail modes, turning from a safeguard against collapse into a mechanism that precipitates it. We theoretically prove that such siloed selection accelerates collapse and induces power-law diversity decay. As an initial mitigation, we construct Wasserstein proxy references from multiple silos without sharing raw data. Empirical results confirm that local-reference selection fails on skewed distributions, whereas collaborative proxy references mitigate diversity degradation, suggesting that recursive synthetic-data pipelines require particular caution when real-data coverage is fragmented or scarce.

Problem

• sample selection bias
• model collapse
• data silos
• synthetic data
• distributional tails

Innovation

• model collapse
• sample selection bias
• data silos
• Wasserstein proxy
• synthetic data