DELTA: DEep Learning Transfer using Feature Map with Attention for Convolutional Networks
This addresses the problem of limited dataset size in transfer learning for practitioners, though it is incremental as it builds on existing regularization methods.
The paper tackles the accuracy bottleneck in transfer learning when fine-tuning pre-trained networks on small target datasets by proposing DELTA, a framework that aligns outer layer outputs using attention-selected feature maps, achieving higher accuracy than L2 and L2-SP baselines.
Transfer learning through fine-tuning a pre-trained neural network with an extremely large dataset, such as ImageNet, can significantly accelerate training while the accuracy is frequently bottlenecked by the limited dataset size of the new target task. To solve the problem, some regularization methods, constraining the outer layer weights of the target network using the starting point as references (SPAR), have been studied. In this paper, we propose a novel regularized transfer learning framework DELTA, namely DEep Learning Transfer using Feature Map with Attention. Instead of constraining the weights of neural network, DELTA aims to preserve the outer layer outputs of the target network. Specifically, in addition to minimizing the empirical loss, DELTA intends to align the outer layer outputs of two networks, through constraining a subset of feature maps that are precisely selected by attention that has been learned in an supervised learning manner. We evaluate DELTA with the state-of-the-art algorithms, including L2 and L2-SP. The experiment results show that our proposed method outperforms these baselines with higher accuracy for new tasks.