mlsanityData to 14 Oct 2026

Explaining How Quantization Disparately Skews a Model

arXiv 2509.07222 ↗First posted 8 Sep 2025cs.LG cs.AI cs.CYcited by 2

Not on the map. Why?

Abstract (arXiv)

Post Training Quantization (PTQ) is widely adopted due to its high compression capacity and speed with minimal impact on accuracy. However, we observed that disparate impacts are exacerbated by quantization, especially for minority groups. Our analysis explains that in the course of quantization there is a chain of factors attributed to a disparate impact across groups during forward and backward passes. We explore how the changes in weights and activations induced by quantization cause cascaded impacts in the network, resulting in logits with lower variance, increased loss, and compromised group accuracies. We extend our study to verify the influence of these impacts on group gradient norms and eigenvalues of the Hessian matrix, providing insights into the state of the network from an optimization point of view. To mitigate these effects, we propose integrating mixed precision Quantization Aware Training (QAT) with dataset sampling methods and weighted loss functions, therefore providing fair deployment of quantized neural networks.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 32 of its 42 references that are in the map. First posted on arXiv 8 Sep 2025.

Builds on

32 of 42 references.

References · 32

  1. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,329
  2. 2018The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networkscited by 2,262
  3. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 2,031
  4. 2016Equality of Opportunity in Supervised Learningcited by 2,007
  5. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,919
  6. 2016On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minimacited by 1,352
  7. 2017Visualizing the Loss Landscape of Neural Netscited by 1,130
  8. 2019Learned Step Size Quantizationcited by 628
  9. 2017To prune, or not to prune: exploring the efficacy of pruning for model compressioncited by 603
  10. 2021BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstructioncited by 479
  11. 2021A White Paper on Neural Network Quantizationcited by 472
  12. 2017Age Progression/Regression by Conditional Adversarial Autoencodercited by 429
20 more
  1. 2016Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activationscited by 426
  2. 2019HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precisioncited by 402
  3. 2022Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruningcited by 322
  4. 2022The case for 4-bit precision: k-bit Inference Scaling Lawscited by 251
  5. 2019Low-bit Quantization of Neural Networks for Efficient Inferencecited by 193
  6. 2020LSQ+: Improving low-bit quantization through learnable offsets and better initializationcited by 189
  7. 2019What Do Compressed Deep Neural Networks Forget?cited by 171
  8. 2019Improving Neural Network Quantization without Retraining using Outlier Channel Splittingcited by 151
  9. 2019Inherent Tradeoffs in Learning Fair Representationscited by 144
  10. 2020To be Robust or to be Fair: Towards Fairness in Adversarial Trainingcited by 137
  11. 2018Mixed Precision Quantization of ConvNets via Differentiable Neural Architecture Searchcited by 125
  12. 2022Penalizing Gradient Norm for Efficiently Improving Generalization in Deep Learningcited by 122
  13. 2022Overcoming Oscillations in Quantization-Aware Trainingcited by 119
  14. 2023Memory-Efficient Fine-Tuning of Compressed Large Language Models via sub-4-bit Integer Quantizationcited by 100
  15. 2018Achieving Fairness through Adversarial Learning: an Application to Recidivism Predictioncited by 77
  16. 2021Fairness via Representation Neutralizationcited by 63
  17. 2023FlexRound: Learnable Rounding based on Element-wise Division for Post-Training Quantizationcited by 48
  18. 2023Pruning vs Quantization: Which is Better?cited by 46
  19. 2022Pruning has a disparate impact on model accuracycited by 44
  20. 2023Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluationcited by 25

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Decoupling Generalizability and Membership Privacy Risks in Neural Networkscited by 1
  2. 2026Which Decisions Low-Bit Quantization Breaks, and How to Predict Themcited by 1

Similar, not cited

Close in wording, no citation either way

  1. 2025Beyond Outliers: A Study of Optimizers Under Quantization
  2. 2023MetaGrad: Adaptive Gradient Quantization with Hypernetworks
  3. 2024Art and Science of Quantizing Large-Scale Models: A Comprehensive Overview
  4. 2026Investigating Social Bias Changes in Quantized Language Models
  5. 2023Augmenting Hessians with Inter-Layer Dependencies for Mixed-Precision Post-Training Quantization
  6. 2026Activation Sensitivity as a Unifying Principle for Post-Training Quantization
  7. 2021Cluster-Promoting Quantization with Bit-Drop for Minimizing Network Quantization Loss
  8. 2024Investigating the Impact of Quantization on Adversarial Robustness