LoRA Subtraction for Drift-Resistant Space in Exemplar-Free Continual Learning

2025-03-23

AI Summary

To address catastrophic forgetting caused by feature drift in exemplar-free continual learning (EFCL), this paper proposes the Drift-Resistant Space (DRS)—a novel framework requiring neither stored old-task data nor statistical summaries. Methodologically, DRS introduces the first LoRA-weight subtraction mechanism to dynamically recalibrate pretrained model parameters, ensuring stable alignment between old and new task feature spaces. It abandons static modeling and historical caching, instead jointly optimizing stability and plasticity via triplet loss. Fully compatible with parameter-efficient fine-tuning paradigms, DRS achieves state-of-the-art performance across long task sequences (≥10 tasks) on multiple benchmarks. Empirical results show consistent average accuracy improvements of 3.2–5.7% over prior EFCL methods, demonstrating substantial mitigation of forgetting without exemplars or memory buffers.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

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

In continual learning (CL), catastrophic forgetting often arises due to feature drift. This challenge is particularly prominent in the exemplar-free continual learning (EFCL) setting, where samples from previous tasks cannot be retained, making it difficult to preserve prior knowledge. To address this issue, some EFCL methods aim to identify feature spaces that minimize the impact on previous tasks while accommodating new ones. However, they rely on static features or outdated statistics stored from old tasks, which prevents them from capturing the dynamic evolution of the feature space in CL, leading to performance degradation over time. In this paper, we introduce the Drift-Resistant Space (DRS), which effectively handles feature drifts without requiring explicit feature modeling or the storage of previous tasks. A novel parameter-efficient fine-tuning approach called Low-Rank Adaptation Subtraction (LoRA-) is proposed to develop the DRS. This method subtracts the LoRA weights of old tasks from the initial pre-trained weight before processing new task data to establish the DRS for model training. Therefore, LoRA- enhances stability, improves efficiency, and simplifies implementation. Furthermore, stabilizing feature drifts allows for better plasticity by learning with a triplet loss. Our method consistently achieves state-of-the-art results, especially for long task sequences, across multiple datasets.

Problem

• Addresses catastrophic forgetting in continual learning
• Handles feature drift without storing old tasks
• Improves stability and efficiency in model training

Innovation

• Drift-Resistant Space handles feature drifts dynamically
• LoRA Subtraction fine-tunes weights for stability
• Triplet loss enhances plasticity in learning