Dyn-Adapter：面向高效视觉识别的解耦表示
Dyn-Adapter: Towards Disentangled Representation for Efficient Visual Recognition
AI总结：
提出Dyn-adapter，通过多层级特征解耦的动态架构与双向稀疏化策略，在维持或提升识别准确率的同时将推理FLOPs减少50%，有效缓解了参数高效迁移学习的计算复杂度与推理负担。
摘要：
参数高效迁移学习（PETL）是一项极具前景的任务，旨在以相对适中的成本将大规模预训练模型适配至下游任务。然而，当前的PETL方法在压缩计算复杂度方面存在困难，且由于完整的前向过程而承受沉重的推理负担。本文提出了一种高效的视觉识别范式，称为Dynamic Adapter（Dyn-adapter），通过巧妙地解耦多层级特征来提升PETL效率。我们的方法很简单：首先，我们设计了一种带有平衡早期头部（early heads）的动态架构，用于多层级特征提取，并配合自适应训练策略。其次，我们引入了一种由追求强大泛化能力所驱动的双向稀疏化策略。这些特性使我们能够高效且有效地进行微调：我们在推理过程中将FLOPs减少了50%，同时维持甚至获得了更高的识别准确率。在多样化数据集和预训练骨干网络上的大量实验表明，Dyn-adapter作为视觉识别任务中PETL的通用效率提升器具有巨大潜力。
英文摘要：
Parameter-efficient transfer learning (PETL) is a promising task, aiming to adapt the large-scale pre-trained model to downstream tasks with a relatively modest cost. However, current PETL methods struggle in compressing computational complexity and bear a heavy inference burden due to the complete forward process. This paper presents an efficient visual recognition paradigm, called Dynamic Adapter (Dyn-Adapter), that boosts PETL efficiency by subtly disentangling features in multiple levels. Our approach is simple: first, we devise a dynamic architecture with balanced early heads for multi-level feature extraction, along with adaptive training strategy. Second, we introduce a bidirectional sparsity strategy driven by the pursuit of powerful generalization ability. These qualities enable us to fine-tune efficiently and effectively: we reduce FLOPs during inference by 50%, while maintaining or even yielding higher recognition accuracy. Extensive experiments on diverse datasets and pretrained backbones demonstrate the potential of Dyn-Adapter serving as a general efficiency booster for PETL in vision recognition tasks.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
