Progressive Prompts: Continual Learning for Language Models

By Anastasia Razdaibiedina, Yuning Mao, Rui Hou, Madian Khabsa, Michael Lewis, Amjad Almahairi. 2023

243 citations recorded by the source corpus.

We introduce Progressive Prompts - a simple and efficient approach for continual learning in language models. Our method allows forward transfer and resists catastrophic forgetting, without relying on data replay or a large number of task-specific parameters. Progressive Prompts learns a new soft prompt for each task and sequentially concatenates it with the previously learned prompts, while keeping the base model frozen. Experiments on standard continual learning benchmarks show that our approach outperforms state-of-the-art methods, with an improvement>20% in average test accuracy over the previous best-preforming method on T5 model. We also explore a more challenging continual learning setup with longer sequences of tasks and show that Progressive Prompts significantly outperforms prior methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Large Language Models, Prompting & Reasoning, Meta & Continual Learning

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. Hang Su
2. Jun Zhu
3. Jorge Armando Mendez Mendez
4. Eric Eaton
5. Liyuan Wang
6. Xingxing Zhang
7. Vinay Kumar Verma
8. Xuanjing Huang

In Computer Vision and Pattern Recognition

1. Hang Su
2. Jun Zhu
3. Liyuan Wang
4. Xingxing Zhang
5. Jorge Armando Mendez Mendez
6. Eric Eaton
7. Vinay Kumar Verma
8. Xuanjing Huang

In Large Language Models

1. Jorge Armando Mendez Mendez
2. Liyuan Wang
3. Eric Eaton
4. Jun Zhu
5. Xuanjing Huang
6. Hang Su
7. Xingxing Zhang

In Prompting & Reasoning

1. Liyuan Wang
2. Jun Zhu
3. Xuanjing Huang
4. Hang Su
5. Xingxing Zhang

In Meta & Continual Learning

1. Jorge Armando Mendez Mendez
2. Eric Eaton
3. Xingxing Zhang
4. Liyuan Wang
5. Hang Su
6. Jun Zhu
7. Vinay Kumar Verma
8. Bing Liu

Citations in the corpus

• A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “…optimizing a weighted summation of the prompt pool with attention factors (CODA-Prompt [206]), using explicitly task-sharing and task-specific prompts (DualPrompt [207]) or only task-specific prompts (S-Prompts [208]), progressive expansion of task-specific prompts (Progressive Prompts [209]), etc.”
• Convolutional Prompting meets Language Models for Continual Learning (2024, background): “ProgressivePrompts [37] progressively learns new prompt tokens for each incoming tasks, but assumes the presence of task-id during inference.”
• Consistent Prompting for Rehearsal-Free Continual Learning (2024, unknown): “Recent prompt-based continual learning methods [14, 29, 33] handle the rehearsal-free setting by encoding task-specific knowledge within prompts.”
• A Comprehensive Overview of Large Language Models (2023, unknown): “In another study [189], the authors enable LLMs to reason over sources of natural language feedback, forming an “inner monologue” that enhances their ability to process and plan actions in robotic control scenarios.”
• Continual Segmentation with Disentangled Objectness Learning and Class Recognition (2024, unknown): “Recently, more works [13, 37, 41, 41, 49, 50] started to leverage the dynamic nature of Transformers.”
• How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Composition (2022, background): “A more recent mechanism that follows this same direction learns a continuous input vector per task to adapt the behavior of a large pretrained language model (Razdaibiedina et al., 2023).”
• Orthogonal Subspace Learning for Language Model Continual Learning (2023, methodology): “• ProgPrompt (Razdaibiedina et al., 2023): adopts a task-specific soft prompt for each distinct task, sequentially appending it to prior learned prompts.”