LGAIJul 26, 2021

Don't Sweep your Learning Rate under the Rug: A Closer Look at Cross-modal Transfer of Pretrained Transformers

arXiv:2107.12460v16 citations
Originality Synthesis-oriented
AI Analysis

This work corrects a methodological oversight in cross-modal transfer research, emphasizing hyperparameter tuning for robust findings.

The authors re-evaluated a prior claim that frozen pretrained transformers match or outperform fine-tuned ones in cross-modal transfer tasks, finding it was due to improper learning rate tuning. After proper tuning, they showed pretrained transformers outperform training from scratch only when fully fine-tuned.

Self-supervised pre-training of large-scale transformer models on text corpora followed by finetuning has achieved state-of-the-art on a number of natural language processing tasks. Recently, Lu et al. (2021, arXiv:2103.05247) claimed that frozen pretrained transformers (FPTs) match or outperform training from scratch as well as unfrozen (fine-tuned) pretrained transformers in a set of transfer tasks to other modalities. In our work, we find that this result is, in fact, an artifact of not tuning the learning rates. After carefully redesigning the empirical setup, we find that when tuning learning rates properly, pretrained transformers do outperform or match training from scratch in all of our tasks, but only as long as the entire model is finetuned. Thus, while transfer from pretrained language models to other modalities does indeed provide gains and hints at exciting possibilities for future work, properly tuning hyperparameters is important for arriving at robust findings.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes