Explicit Inductive Bias for Transfer Learning with Convolutional Networks

Abstract

In inductive transfer learning, fine-tuning pre-trained convolutional networks substantially outperforms training from scratch. When using fine-tuning, the underlying assumption is that the pre-trained model extracts generic features, which are at least partially relevant for solving the target task, but would be difficult to extract from the limited amount of data available on the target task. However, besides the initialization with the pre-trained model and the early stopping, there is no mechanism in fine-tuning for retaining the features learned on the source task. In this paper, we investigate several regularization schemes that explicitly promote the similarity of the final solution with the initial model. We show the benefit of having an explicit inductive bias towards the initial model, and we eventually recommend a simple $L^2$ penalty with the pre-trained model being a reference as the baseline of penalty for transfer learning tasks.

arXiv commentAccepted at ICML 2018

Cited by in corpus (11)

• Rethinking the Hyperparameters for Fine-tuning
• Attend and Guide (AG-Net): A Keypoints-driven Attention-based Deep Network for Image Recognition
• SpotTune: Transfer Learning through Adaptive Fine-tuning
• A Comprehensive Review and a Taxonomy of Edge Machine Learning: Requirements, Paradigms, and Techniques
• Distribution-aware Margin Calibration for Semantic Segmentation in Images
• Self Attention with Temporal Prior: Can We Learn More from Arrow of Time?
• Representation Transfer by Optimal Transport
• Transfer Learning with Pre-trained Conditional Generative Models
• Transfer Reinforcement Learning across Homotopy Classes
• StochCA: A Novel Approach for Exploiting Pretrained Models with Cross-Attention
• Towards Making Deep Transfer Learning Never Hurt