# Learning Stable Predictors from Weak Supervision under Distribution Shift

arXiv: https://arxiv.org/abs/2604.05002 · HTML: https://mlsanity.com/paper/2604.05002/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 20 references that are in the map.
- First posted on arXiv 5 Apr 2026.

## Builds on (most cited first, 12 of 12)

- [Invariant Risk Minimization](https://mlsanity.com/paper/1907.02893/) (2019)
- [Causal inference using invariant prediction: identification and confidence intervals](https://mlsanity.com/paper/1501.01332/) (2015)
- [Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision](https://mlsanity.com/paper/2312.09390/) (2023)
- [Training Complex Models with Multi-Task Weak Supervision](https://mlsanity.com/paper/1810.02840/) (2018)
- [Universalizing Weak Supervision](https://mlsanity.com/paper/2112.03865/) (2021)
- [Pairwise Supervision Can Provably Elicit a Decision Boundary](https://mlsanity.com/paper/2006.06207/) (2020)
- [Looking Beyond Label Noise: Shifted Label Distribution Matters in Distantly Supervised Relation Extraction](https://mlsanity.com/paper/1904.09331/) (2019)
- [Learnability with Indirect Supervision Signals](https://mlsanity.com/paper/2006.08791/) (2020)
- [A General Framework for Learning from Weak Supervision](https://mlsanity.com/paper/2402.01922/) (2024)
- [Weak-to-Strong Generalization under Distribution Shifts](https://mlsanity.com/paper/2510.21332/) (2025)
- [Learning from Noisy Similar and Dissimilar Data](https://mlsanity.com/paper/2002.00995/) (2020)
- [XPASC: Measuring Generalization in Weak Supervision by Explainability and Association](https://mlsanity.com/paper/2206.01444/) (2022)

## Built on by (most cited first, 1 of 1)

- [A Decision-Support Audit Protocol for Supervision Drift in Proxy-Labeled Credit-Risk Prediction](https://mlsanity.com/paper/2609.16102/) (2026)

## Abstract (arXiv)

Learning from weak, proxy, or relative supervision is common when ground-truth labels are unavailable, but robustness under distribution shift remains poorly understood because the supervision mechanism itself may change across environments. We formalize this phenomenon as supervision drift, defined as changes in P(y mid x, c) across contexts, and study it in CRISPR-Cas13d transcriptomic perturbation experiments where guide efficacy is inferred indirectly from RNA-seq responses. Using publicly available data spanning two human cell lines and multiple post-induction timepoints, we construct a controlled non-IID benchmark with explicit domain (cell line) and temporal shifts, while reusing a fixed weak-label construction across all contexts to avoid changing targets. Across linear and tree-based models, weak supervision supports meaningful learning in-domain (ridge R2 = 0.356, Spearman ρ = 0.442) and partial cross-cell-line transfer (ρ ≈ 0.40). In contrast, temporal transfer collapses across all model classes considered, yielding negative R2 and weak or near-zero ρ (ridge R2 = -0.145, ρ = 0.008; XGBoost R2 = -0.155, ρ = 0.056; random forest R2 = -0.322, ρ = 0.139). Additional robustness analyses using externally recomputed weak labels, shift-score quantification, and simple mitigation baselines preserve the same qualitative pattern. Feature-label association and feature-importance analyses remain relatively stable across cell lines but change sharply over time, indicating that failures arise from supervision drift rather than model capacity or simple covariate shift. These results show that strong in-domain performance under weak supervision can be misleading and motivate feature stability as a lightweight diagnostic for non-transferability before deployment.