Toward Multi-Domain and Long-Tailed Quantization via Feature Alignment and Scaling
2026-06-03 • Machine Learning
Machine LearningComputer Vision and Pattern Recognition
AI summaryⓘ
The authors address the problem of making deep neural networks work well on small devices when the data comes from different sources or has imbalanced classes. They propose a method called EmaQ that helps the network handle multiple data domains better by aligning distributions and carefully combining weights. For datasets with many rare classes, they extend this to EmaQ-LT, which adjusts for overconfidence in common classes. Their theory supports the method's stability, and experiments show good performance in these challenging situations.
deep neural networksquantizationdomain shiftclass imbalanceCDF projectionweight aggregationlong-tailed distributionvariance scalinglogit adjustmentconvergence guarantees
Authors
Chin-Yuan Yeh, Ting-An Chen, De-Nian Yang, Ming-Syan Chen
Abstract
Quantizing deep neural networks is essential for efficient inference on resource-constrained devices. However, most existing methods are designed for single-domain and class-balanced data, leaving practical settings with domain shifts or severe class imbalance underexplored. We address these challenges with Efficient Multi-Domain Alignment Quantization (EmaQ), which aligns domain distributions through a CDF-based projection and uses sensitivity-aware weight aggregation to stabilize multi-domain quantization. We further extend EmaQ to EmaQ-LT for long-tailed quantization by introducing class-conditioned variance scaling and confidence-based logit adjustment to mitigate majority-class overconfidence. Theoretical analyses establish convergence guarantees and motivate the proposed sensitivity and scaling mechanisms. Experiments on standard, multi-domain (Office-31, Digits), and long-tailed (SynDigits-LT, CIFAR-10-LT, CIFAR-100-LT) benchmarks show that EmaQ and EmaQ-LT achieve strong low-bit performance under domain shift and class imbalance.