Why Do Pretrained Language Models Help in Downstream Tasks? An Analysis of Head and Prompt Tuning

By Colin Wei, Sang Michael Xie, Tengyu Ma. 2021

119 citations recorded by the source corpus.

Pretrained language models have achieved state-of-the-art performance when adapted to a downstream NLP task. However, theoretical analysis of these models is scarce and challenging since the pretraining and downstream tasks can be very different. We propose an analysis framework that links the pretraining and downstream tasks with an underlying latent variable generative model of text -- the downstream classifier must recover a function of the posterior distribution over the latent variables. We analyze head tuning (learning a classifier on top of the frozen pretrained model) and prompt tuning in this setting. The generative model in our analysis is either a Hidden Markov Model (HMM) or an HMM augmented with a latent memory component, motivated by long-term dependencies in natural language. We show that 1) under certain non-degeneracy conditions on the HMM, simple classification heads can solve the downstream task, 2) prompt tuning obtains downstream guarantees with weaker non-degeneracy conditions, and 3) our recovery guarantees for the memory-augmented HMM are stronger than for the vanilla HMM because task-relevant information is easier to recover from the long-term memory. Experiments on synthetically generated data from HMMs back our theoretical findings.

Subfields: Artificial Intelligence, Large Language Models, Prompting & Reasoning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Ashish Sabharwal
2. Zhiyuan Liu
3. Peter Clark
4. Tushar Khot
5. Maosong Sun
6. Hannaneh Hajishirzi
7. Ning Ding
8. Fei Huang

In Large Language Models

1. Maosong Sun
2. Zhiyuan Liu
3. Fei Huang
4. Shumin Deng
5. Shengding Hu
6. Haitao Zheng
7. Chuanqi Tan
8. Weilin Zhao

In Prompting & Reasoning

1. Peter Clark
2. Ashish Sabharwal
3. Tushar Khot
4. Zhiyuan Liu
5. Maosong Sun
6. Ning Ding
7. Fei Huang
8. Shumin Deng

Citations in the corpus

• Emergent Abilities of Large Language Models (2022, neutral): “Moreover, better exploration of what makes prompting successful (Wei et al., 2021a; Xie et al., 2022; Min et al., 2022; Olsson et al., 2022) could lead to insights on how to elicit emergent abilities at a smaller model scale.”
• An Explanation of In-context Learning as Implicit Bayesian Inference (2021, neutral): “Note that exp( n · r n ( θ ∗ ))=1 for θ ∗ .”
• OpenPrompt: An Open-source Framework for Prompt-learning (2021, background): “These practical studies imply that we may use prompts to more effectively and efficiently dig the knowledge kept in PLMs, leading to a deeper understanding of the underlying principles of their mechanisms (Wei et al., 2021; Qin et al., 2021; Vu et al., 2021).”
• Complexity-Based Prompting for Multi-Step Reasoning (2022, background): “There are further works linking the relationship between in-context learning and classical Bayesian inference (Wei et al., 2021; Xie et al., 2022).”
• Reasoning with Language Model Prompting: A Survey (2022, background): “Some works concentrate on unraveling the principles of machine learning algorithm behind in-context learning (Xie et al., 2022; Akyürek et al., 2023) and prompt tuning (Wei et al., 2021).”