Are Pre-trained Convolutions Better than Pre-trained Transformers?
This work addresses the problem of architectural bias in pre-trained language models for NLP researchers, suggesting that pre-training and architecture should be considered separately, though it is incremental as it builds on existing CNN and Transformer research.
The paper investigates whether convolutional neural network (CNN) pre-trained language models can compete with Transformer-based ones, finding that CNNs are competitive and outperform Transformers in some scenarios across 8 datasets/tasks, with caveats.
In the era of pre-trained language models, Transformers are the de facto choice of model architectures. While recent research has shown promise in entirely convolutional, or CNN, architectures, they have not been explored using the pre-train-fine-tune paradigm. In the context of language models, are convolutional models competitive to Transformers when pre-trained? This paper investigates this research question and presents several interesting findings. Across an extensive set of experiments on 8 datasets/tasks, we find that CNN-based pre-trained models are competitive and outperform their Transformer counterpart in certain scenarios, albeit with caveats. Overall, the findings outlined in this paper suggest that conflating pre-training and architectural advances is misguided and that both advances should be considered independently. We believe our research paves the way for a healthy amount of optimism in alternative architectures.