Are Pre-trained Convolutions Better than Pre-trained Transformers?

Abstract

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.

arXiv commentACL'21 + updated code/ckpt pointers

References in corpus (16)

• Sequence to Sequence Learning with Neural Networks
• An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling
• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• Character-level Convolutional Networks for Text Classification
• Language Models are Few-Shot Learners
• Convolutional Sequence to Sequence Learning
• Pre-trained Models for Natural Language Processing: A Survey
• SQuAD: 100,000+ Questions for Machine Comprehension of Text
• MASS: Masked Sequence to Sequence Pre-training for Language Generation
• Pay Less Attention with Lightweight and Dynamic Convolutions
• Depthwise Separable Convolutions for Neural Machine Translation
• Synthesizer: Rethinking Self-Attention in Transformer Models
• Adafactor: Adaptive Learning Rates with Sublinear Memory Cost
• StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding
• Mesh-TensorFlow: Deep Learning for Supercomputers
• COGS: A Compositional Generalization Challenge Based on Semantic Interpretation

Cited by in corpus (7)

• Do Vision Transformers See Like Convolutional Neural Networks?
• Charformer: Fast Character Transformers via Gradient-based Subword Tokenization
• Scale Efficiently: Insights from Pre-training and Fine-tuning Transformers
• ReduNet: A White-box Deep Network from the Principle of Maximizing Rate Reduction
• Emergency Vehicles Audio Detection and Localization in Autonomous Driving
• The Benchmark Lottery
• Compositional generalization in semantic parsing with pretrained transformers