SAAIL Lab SAAIL Lab
ICLR 2026 Workshop 2+ Citations Accepted & Published · ICLR 2026 Workshop

Unifying Perspectives on Learning Biases: A Data-Centric Intervention for Holistic Fairness, Robustness, and Generalization

1School of Engineering and Science, Indian Institute of Technology Madras Zanzibar, Tanzania
2Wadhwani School of Data Science and AI (WSDSAI), Indian Institute of Technology Madras, Chennai, India
ICLR 2026 Workshop on Principled Design for Trustworthy AI: Interpretability, Robustness, and Safety across Modalities, Rio de Janeiro, Brazil (Apr 2026)
Unifying Perspectives on Learning Biases: A Data-Centric Intervention for Holistic Fairness, Robustness, and Generalization

Summary: Accepted paper at the ICLR 2026 Trustworthy AI Workshop unifying fairness, robustness, and out-of-distribution generalization via data-centric bias intervention.

Abstract

Fairness, robustness, and out-of-distribution generalization are typically studied in isolation despite sharing common roots in dataset acquisition bias, spurious shortcuts, and representational imbalance. We propose a unified data-centric framework that diagnoses and intervenes on learning biases prior to model fitting, demonstrating simultaneous gains across fairness disparities, environmental perturbations, and cross-domain transfer in resource-constrained deployments.

Keywords: trustworthy AIfairnessrobustnessgeneralizationdata-centric AIICLR 2026learning biases

Key Findings & Contributions

22
Models Audited
CNNs, ViTs, Decomposed
99.45%
Decomposed F1
100.0% EER (U-Net + ResNet-18)
97.2%
Shifted Field F1
vs 31.2-45.8% monolithic
5,898
Expert-Graded Images
3 certified ZSTC graders
  1. Formalizes procedural alignment, structuring deep learning pipelines to mirror explicit human segmentation-and-attribute grading stages.
  2. Evaluates 22 architectures across five model families, demonstrating that high clean-set accuracy (99.71%) coincides with negligible explanation faithfulness (2.5-5.8% EER).
  3. Exposes severe domain-shift vulnerability in monolithic black box classifiers (F1 dropping to 31.2-45.8%) compared with 97.2% retained F1 in decomposed architectures.
  4. Unifies fairness, out-of-distribution robustness, and explainability as linked consequences of background shortcut bias.

Benchmark Results

Accuracy, Explanation Evidence Ratio (EER), and Domain-Shift Resilience (ICLR 2026)

ArchitectureClean F1 (%)EER (%)Shifted F1 (%)Params (M)
U-Net + ResNet-18 (Decomposed)99.45100.097.213.2
Swin-Base (Monolithic)99.573.7 ± 3.245.888.0
DeiT-Base (Monolithic)99.574.2 ± 3.642.186.6
DenseNet-201 (Monolithic)99.713.1 ± 2.838.720.0
EfficientNet-B2 (Monolithic)99.712.8 ± 3.434.59.1
ResNet-50 (Monolithic)99.712.5 ± 3.131.225.6

BibTeX

@inproceedings{vincent2026unifying,
  title={Unifying Perspectives on Learning Biases: A Data-Centric Intervention for Holistic Fairness, Robustness, and Generalization},
  author={Vincent, Patrick Ndowo and Nyalala, Innocent},
  booktitle={ICLR 2026 Workshop on Principled Design for Trustworthy AI: Interpretability, Robustness, and Safety across Modalities},
  address={Rio de Janeiro, Brazil},
  year={2026},
  url={https://openreview.net/forum?id=RCEEHTbSDk}
}
APA Citation

Vincent, P. N., & Nyalala, I. (2026). Unifying Perspectives on Learning Biases: A Data-Centric Intervention for Holistic Fairness, Robustness, and Generalization. In ICLR 2026 Workshop on Principled Design for Trustworthy AI, Rio de Janeiro. https://openreview.net/forum?id=RCEEHTbSDk