QuAIL: Quality-Aware Inertial Learning for Robust Training under Data Corruption

Tabular machine learning systems are frequently trained on data affected by non-uniform corruption, including noisy measurements, missing entries, and feature-specific biases. In practice, these defects are often documented only through column-level reliability indicators rather than instance-wise quality annotations, limiting the applicability of many robustness and cleaning techniques. We present QuAIL, a quality-informed training mechanism that incorporates feature reliability priors directly into the learning process. QuAIL augments existing models with a learnable feature-modulation layer whose updates are selectively constrained by a quality-dependent proximal regularizer, thereby inducing controlled adaptation across features of varying trustworthiness. This stabilizes optimization under structured corruption without explicit data repair or sample-level reweighting. Empirical evaluation across 50 classification and regression datasets demonstrates that QuAIL consistently improves average performance over neural baselines under both random and value-dependent corruption, with especially robust behavior in low-data and systematically biased settings. These results suggest that incorporating feature reliability information directly into optimization dynamics is a practical and effective approach for resilient tabular learning.

💡 Research Summary

QuAIL (Quality‑Aware Inertial Learning) addresses a common yet under‑explored problem in tabular machine learning: training on data that suffers from non‑uniform corruption such as noisy measurements, missing entries, and feature‑specific biases, where only column‑level reliability indicators are available. Traditional robustness techniques often rely on instance‑level quality labels or costly pre‑processing pipelines, making them unsuitable for many production settings. QuAIL instead embeds feature reliability priors directly into the optimization dynamics of any differentiable tabular model.

The method augments a base predictor fθ with a learnable diagonal gating layer g ∈ ℝ^D. For each input sample xi, the gated representation \tilde xi = g ⊙ xi scales each feature by its gate value. The gates are trained jointly with the model parameters using a composite loss L = (1/N)∑i Ltask(ŷi, yi) + λ Lgate. The gate regularizer Lgate is a quality‑weighted proximal term: Lgate = (1/D)∑j wj (gj − ganchor,j)^2, where wj = ϕ(1 − qj) and qj ∈