AdaFilter: Adaptive Filter Fine-tuning for Deep Transfer Learning
This addresses the challenge of transfer learning inefficiencies for researchers and practitioners in computer vision, offering an incremental improvement over existing fine-tuning methods.
The paper tackles the problem of effectively using pre-trained deep neural networks for new tasks by proposing AdaFilter, an adaptive fine-tuning approach that selects and optimizes only a subset of convolutional filters per example, reducing the average classification error by 2.54% compared to standard fine-tuning across 7 image classification datasets.
There is an increasing number of pre-trained deep neural network models. However, it is still unclear how to effectively use these models for a new task. Transfer learning, which aims to transfer knowledge from source tasks to a target task, is an effective solution to this problem. Fine-tuning is a popular transfer learning technique for deep neural networks where a few rounds of training are applied to the parameters of a pre-trained model to adapt them to a new task. Despite its popularity, in this paper, we show that fine-tuning suffers from several drawbacks. We propose an adaptive fine-tuning approach, called AdaFilter, which selects only a part of the convolutional filters in the pre-trained model to optimize on a per-example basis. We use a recurrent gated network to selectively fine-tune convolutional filters based on the activations of the previous layer. We experiment with 7 public image classification datasets and the results show that AdaFilter can reduce the average classification error of the standard fine-tuning by 2.54%.