When majority rules, minority loses: bias amplification of gradient descent
cs.LG, cs.AI, math.OC
Submitted: 2025-05-19
Updated: 2026-09-15
Journal ref: NeurIPS 2025 - Thirty-ninth Annual Conference on Neural Information Processing Systems, Dec 2025, San Diego (CA), United States
Code: https://github.com/ryanboustany/bias_amplification
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- How regularization affects the geometry of loss functions
- Review of Mathematical frameworks for Fairness in Machine Learning
- Gradient Descent Finds Global Minima of Deep Neural Networks
- ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
- A Systematic Study of Bias Amplification
- SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
- Exploring the ultra-light to sub-MeV dark matter window with atomic clocks and co-magnetometers
- An Effective Theory of Bias Amplification
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks