Restoring Neural Network Plasticity for Faster Transfer Learning
This addresses the issue of inefficient transfer learning for computer vision practitioners, but it is incremental as it builds on known plasticity problems in continual learning.
The paper tackles the problem of loss of neural plasticity in transfer learning, where pretrained weights can hinder adaptation to downstream tasks, by proposing a targeted weight re-initialization strategy. The result shows that this method improves test accuracy and convergence speed for CNNs and ViTs on image classification benchmarks with negligible computational overhead.
Transfer learning with models pretrained on ImageNet has become a standard practice in computer vision. Transfer learning refers to fine-tuning pretrained weights of a neural network on a downstream task, typically unrelated to ImageNet. However, pretrained weights can become saturated and may yield insignificant gradients, failing to adapt to the downstream task. This hinders the ability of the model to train effectively, and is commonly referred to as loss of neural plasticity. Loss of plasticity may prevent the model from fully adapting to the target domain, especially when the downstream dataset is atypical in nature. While this issue has been widely explored in continual learning, it remains relatively understudied in the context of transfer learning. In this work, we propose the use of a targeted weight re-initialization strategy to restore neural plasticity prior to fine-tuning. Our experiments show that both convolutional neural networks (CNNs) and vision transformers (ViTs) benefit from this approach, yielding higher test accuracy with faster convergence on several image classification benchmarks. Our method introduces negligible computational overhead and is compatible with common transfer learning pipelines.