Layer-Wise Evolution of Representations in Fine-Tuned Transformers: Insights from Sparse AutoEncoders
This provides insights into fine-tuning mechanisms for researchers, but it is incremental as it builds on existing understanding of transformer adaptation.
The paper investigates how fine-tuning transforms BERT's representations across layers, finding that early layers retain general features, middle layers transition, and later layers specialize for the task, based on experiments with Sparse AutoEncoders and activation analysis.
Fine-tuning pre-trained transformers is a powerful technique for enhancing the performance of base models on specific tasks. From early applications in models like BERT to fine-tuning Large Language Models (LLMs), this approach has been instrumental in adapting general-purpose architectures for specialized downstream tasks. Understanding the fine-tuning process is crucial for uncovering how transformers adapt to specific objectives, retain general representations, and acquire task-specific features. This paper explores the underlying mechanisms of fine-tuning, specifically in the BERT transformer, by analyzing activation similarity, training Sparse AutoEncoders (SAEs), and visualizing token-level activations across different layers. Based on experiments conducted across multiple datasets and BERT layers, we observe a steady progression in how features adapt to the task at hand: early layers primarily retain general representations, middle layers act as a transition between general and task-specific features, and later layers fully specialize in task adaptation. These findings provide key insights into the inner workings of fine-tuning and its impact on representation learning within transformer architectures.