Understanding Linear Probing then Fine-tuning Language Models from NTK Perspective
Authors:
Akiyoshi Tomihari,
Issei Sato
Abstract:
The two-stage fine-tuning (FT) method, linear probing then fine-tuning (LP-FT), consistently outperforms linear probing (LP) and FT alone in terms of accuracy for both in-distribution (ID) and out-of-distribution (OOD) data. This success is largely attributed to the preservation of pre-trained features, achieved through a near-optimal linear head obtained during LP. However, despite the widespread…
▽ More
The two-stage fine-tuning (FT) method, linear probing then fine-tuning (LP-FT), consistently outperforms linear probing (LP) and FT alone in terms of accuracy for both in-distribution (ID) and out-of-distribution (OOD) data. This success is largely attributed to the preservation of pre-trained features, achieved through a near-optimal linear head obtained during LP. However, despite the widespread use of large language models, the exploration of complex architectures such as Transformers remains limited. In this paper, we analyze the training dynamics of LP-FT for classification models on the basis of the neural tangent kernel (NTK) theory. Our analysis decomposes the NTK matrix into two components, highlighting the importance of the linear head norm alongside the prediction accuracy at the start of the FT stage. We also observe a significant increase in the linear head norm during LP, stemming from training with the cross-entropy (CE) loss, which effectively minimizes feature changes. Furthermore, we find that this increased norm can adversely affect model calibration, a challenge that can be addressed by temperature scaling. Additionally, we extend our analysis with the NTK to the low-rank adaptation (LoRA) method and validate its effectiveness. Our experiments with a Transformer-based model on natural language processing tasks across multiple benchmarks confirm our theoretical analysis and demonstrate the effectiveness of LP-FT in fine-tuning language models. Code is available at https://github.com/tom4649/lp-ft_ntk.
△ Less
Submitted 26 May, 2024;
originally announced May 2024.
