Teaching Prompts to Coordinate: Hierarchical Layer-Grouped Prompt Tuning for Continual Learning

2025-11-15

AI Summary

Existing prompt-based continual learning methods suffer from catastrophic forgetting due to layer-wise independent prompt updates. To address this, we propose Hierarchical Grouped Prompt Tuning (HGPT): model layers are partitioned into groups, with prompts shared within each group; positional encoding is incorporated to preserve feature structural stability. Furthermore, a root-prompt generation mechanism is introduced, wherein a hierarchical network dynamically derives child prompts from a root prompt, enhancing inter-prompt coordination and generalization. HGPT achieves efficient task-specific prompt allocation and cross-task feature alignment without modifying any pre-trained parameters. Evaluated on four standard continual learning benchmarks, HGPT consistently outperforms state-of-the-art prompt-tuning approaches, striking a superior balance between adaptation to new tasks and retention of knowledge from previous tasks.

Technology Category

• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Life-Long and Continual Learning
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Prompt-based continual learning methods fine-tune only a small set of additional learnable parameters while keeping the pre-trained model's parameters frozen. It enables efficient adaptation to new tasks while mitigating the risk of catastrophic forgetting. These methods typically attach one independent task-specific prompt to each layer of pre-trained models to locally modulate its features, ensuring that the layer's representation aligns with the requirements of the new task. However, although introducing learnable prompts independently at each layer provides high flexibility for adapting to new tasks, this overly flexible tuning could make certain layers susceptible to unnecessary updates. As all prompts till the current task are added together as a final prompt for all seen tasks, the model may easily overwrite feature representations essential to previous tasks, which increases the risk of catastrophic forgetting. To address this issue, we propose a novel hierarchical layer-grouped prompt tuning method for continual learning. It improves model stability in two ways: (i) Layers in the same group share roughly the same prompts, which are adjusted by position encoding. This helps preserve the intrinsic feature relationships and propagation pathways of the pre-trained model within each group. (ii) It utilizes a single task-specific root prompt to learn to generate sub-prompts for each layer group. In this way, all sub-prompts are conditioned on the same root prompt, enhancing their synergy and reducing independence. Extensive experiments across four benchmarks demonstrate that our method achieves favorable performance compared with several state-of-the-art methods.

Problem

• Addresses catastrophic forgetting in prompt-based continual learning methods
• Reduces unnecessary layer updates through hierarchical prompt grouping
• Enhances prompt synergy using a shared root prompt structure

Innovation

• Hierarchical layer-grouped prompts coordinate feature updates
• Shared root prompt generates sub-prompts for layer groups
• Position encoding preserves pre-trained model relationships