Accelerated bilevel optimization improves few-shot medical image classification
ABO-Med: Accelerated Bilevel Optimization for Few-Shot Medical Image Classification
Computer Vision and Pattern Recognition
Summary
Medical image classification often struggles when there are only a few examples to learn from. The authors developed ABO-Med, a faster and more scalable optimization approach that improves a popular learning method called MAML. They also created a special image augmentation method designed specifically for medical images, called MedRAug. Their experiments show that ABO-Med makes the classification more accurate, and MedRAug adds additional improvements.
What this means in practice
- •For medical imaging developers: Enhance medical image classification systems to perform better with limited training examples by integrating ABO-Med and tailored augmentation strategies.
- •For machine learning engineers: Build scalable meta-learning solutions capable of fast adaptation through efficient bilevel optimization without second-order derivatives, especially in health-related domains.
Authors
Ruoxuan Shi, Sheng Yang, Zhengxing Su, Xiaoyang Hou, Yating Liu
Abstract
In recent years, bilevel optimization has been widely used in a variety of machine learning tasks. However, prior bilevel optimization algorithms generally require the computation of second-order information, which limits their practical scalability. Only recently has a first-order paradigm for bilevel optimization been established, attaining near-optimal theoretical guarantees for solving bilevel optimization problems. In this paper, we propose ABO-Med, a scalable instantiation of this paradigm for few-shot learning, by incorporating it into the model-agnostic meta-learning (MAML) framework and tailoring it to medical image classification. We also introduce Medical Adaptive RandomAugment (MedRAug), a modality-aware augmentation strategy designed for medical images. Theoretically, ABO-Med establishes the optimality of MAML-type meta-learning approaches. Empirically, ABO-Med outperforms prior baselines on several public medical datasets, with gains of 1.99% to 18.76%, while MedRAug further improves the average accuracy by 2.20% to 6.34%. Additional cross-domain experiments, augmentation ablation studies, backbone ablation studies, and training efficiency analysis further validate the effectiveness and efficiency of the proposed method.