Measuring Task Similarity and Its Implication in Fine-Tuning Graph Neural Networks

• DOI: 10.1609/aaai.v38i11.29156
• Publication Date: 2024-03-25T11:01:39Z
• Authors: Renhong Huang; Jiarong Xu; Xin Jiang; Chenglu Pan; Zhiming Yang; Chunping Wang; Yang Yang
• Keywords: Similarity (geometry); Deep Neural Networks

The paradigm of pre-training and fine-tuning graph neural networks has attracted wide research attention. In previous studies, the pre-trained models are viewed as universally versatile, applied for a diverse range downstream tasks. many situations, however, this practice results in limited or even negative transfer. This paper, first time, emphasizes specific application scope models: not all tasks can effectively benefit from model. light this, we introduce measure task consistency to quantify similarity between assesses extent which Moreover, novel strategy, Bridge-Tune, is proposed further diminish impact difference key innovation Bridge-Tune an intermediate step that bridges takes into account differences refines superiority presented strategy validated via numerous experiments with different