Parameter-Efficient Augment Plugin for Class-Incremental Learning

2025-12-03

AI Summary

To address the dual challenges of the stability-plasticity dilemma and parameter explosion in class-incremental learning (CIL), this paper proposes DLC, a plug-and-play low-rank adaptation paradigm. DLC innovatively integrates LoRA as a lightweight, task-specific residual module directly into deep layers of the base model to enhance task-adaptive representations. A lightweight gating unit dynamically aggregates outputs from multiple task-specific LoRA adapters, effectively mitigating cross-task interference. Crucially, DLC requires no pretraining and is fully compatible with standard CIL strategies such as replay or knowledge distillation. On ImageNet-100, DLC introduces only 4% additional parameters over ResNet-18 yet achieves an 8% absolute accuracy gain; under fixed memory budgets, it significantly outperforms existing state-of-the-art methods. Our core contribution lies in pioneering the plug-and-play, modular application of LoRA in CIL—achieving high accuracy, minimal parameter overhead, and strong scalability across tasks.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Large Vision Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Existing class-incremental learning (CIL) approaches based on replay or knowledge distillation are often constrained by forgetting or the stability-plasticity dilemma. Some expansion-based approaches could achieve higher accuracy. However, they always require significant parameter increases. In this paper, we propose a plugin extension paradigm termed the Deployment of extra LoRA Components (DLC) for non-pre-trained CIL scenarios.We treat the feature extractor trained through replay or distillation as a base model with rich knowledge. For each task, we use Low-Rank Adaptation (LoRA) to inject task-specific residuals into the base model's deep layers. During inference, representations with task-specific residuals are aggregated to produce classification predictions. To mitigate interference from non-target LoRA plugins, we introduce a lightweight weighting unit. This unit learns to assign importance scores to different LoRA-tuned representations. Like downloadable contents in software, our method serves as a plug-and-play enhancement that efficiently extends the base methods. Remarkably, on the large-scale ImageNet-100, with merely 4 % of the parameters of a standard ResNet-18, our DLC model achieves a significant 8 % improvement in accuracy, demonstrating exceptional efficiency. Moreover, it could surpass state-of-the-art methods under the fixed memory budget.

Problem

• Addresses forgetting and stability-plasticity dilemma in class-incremental learning
• Reduces parameter growth in expansion-based incremental learning methods
• Enhances base models with plug-and-play task-specific adaptations efficiently

Innovation

• LoRA components inject task-specific residuals for CIL
• Lightweight weighting unit mitigates interference from non-target plugins
• Plug-and-play enhancement improves accuracy with minimal parameter increase