分离特征提取与分类器优化，显著降低模型训练能耗与时间
Beyond Backbone Backpropagation: A Decoupled Strategy for Efficient Transfer Learning
- 解耦特征提取与分类器训练，预计算特征一次即可
- 在多个模型和医疗数据集上，训练时间大幅减少，精度损失极小
- 适合资源受限的临床或原型开发场景，低碳环保
深度学习模型在图像分类中表现优异，但部署时面临计算成本高、能耗大的挑战。本文提出一种轻量级训练策略：通过适配模型归一化层并解耦特征提取与分类器优化，仅需一次性预计算特征，显著降低开销；重构的分类头采用基于边距的加权损失，避免端到端反向传播带来的模糊性。在四种卷积网络（ResNet18、ResNet50、MobileNet、DenseNet121）、三种视觉变换器（ViT、Swin、DeiT）及三个医学数据集（Brain Cancer MRI、BreakHis、PatchCamelyon）上验证，该方法大幅缩短训练时间，精度仅轻微下降，常达或超过基线水平。效率提升使碳排放降低数个数量级，为资源受限的临床或原型环境提供实用且可持续的解决方案。
原文摘要 · Abstract (English)
Deep learning models achieve state-of-the-art image classification but face deployment challenges due to computational costs and energy demands. We propose a lightweight training strategy that adapts normalization layers of the model to the new domain and decouples feature extraction from classifier optimization, reducing overhead by precomputing features only once. A redesigned classifier head with margin-based weighted loss further minimizes ambiguity without end-to-end backpropagation. Evaluated across four CNN architectures (ResNet18, ResNet50, MobileNet, DenseNet121), three Transformer models (ViT, Swin and DeiT) and three medical datasets (Brain Cancer MRI, BreakHis and PatchCamelyon), our approach significantly reduces the required training time with only a marginal accuracy trade-off, often matching or surpassing baseline performance. This efficiency translates to reducing CO2 by orders of magnitude, offering a practical and environmentally sustainable solution for resource-constrained clinical or prototyping environments.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。