Why Not Hyperparameter-Friendly Optimisation? A Monotonic Adaptive Norm Rescaling Approach For Long-Tailed Recognition

2026-06-01

AI Summary

This work addresses the high sensitivity of classifier retraining to hyperparameters in long-tailed recognition, which severely limits model performance. The authors propose Self-Adaptive Monotonic Normalization (SAMN), a plug-and-play enhancement method that, for the first time, interprets the scaling mechanism of class weight norms from the perspective of class-conditional distributions. SAMN introduces a parameter-free monotonicity constraint strategy based on the Pool Adjacent Violators Algorithm (PAVA), directly enforcing monotonicity without relying on traditional regularization or additional hyperparameters. Extensive experiments on multiple benchmark datasets demonstrate that SAMN consistently achieves significant performance gains, often matching or surpassing state-of-the-art results.

Abstract

Long-tailed recognition poses a significant challenge for deep learning. The two-stage decoupling paradigm, which separates representation learning from classifier retraining, offers a promising solution. During the classifier retraining stage, adaptive norm rescaling is a popular technique. It adjusts the per-class weight norms via parameter regularization, which inevitably introduces hyperparameters. However, many studies report that long-tailed recognition is sensitive to these hyperparameters, as their setup significantly impacts performance. In this paper, we first provide a class-conditional distribution perspective to support norm rescaling methods. Furthermore, we propose a simple but effective approach called Self-Adaptive Monotonic Normalization (SAMN). SAMN avoids the need for parameter regularization. It directly enforces monotonicity on per-class weight norms using the Pool Adjacent Violators Algorithm, making the method hyperparameter-friendly. SAMN is a universal strategy that integrates seamlessly with other methods for enhanced performance. Experiments on benchmark datasets demonstrate that our method significantly boosts long-tailed recognition performance, often achieving state-of-the-art results.

Problem

• long-tailed recognition
• adaptive norm rescaling
• hyperparameter sensitivity
• classifier retraining
• weight norms

Innovation

• long-tailed recognition
• adaptive norm rescaling
• hyperparameter-free
• monotonic normalization
• class-conditional distribution