CKAA: Cross-subspace Knowledge Alignment and Aggregation for Robust Continual Learning

2025-07-12

AI Summary

In parameter-efficient fine-tuning (PEFT) for continual learning, independently trained factorized modules suffer from feature subspace misalignment and ambiguous decision-making under misleading task IDs. To address this, we propose a two-level knowledge alignment and task-confidence-guided adapter mixing mechanism. Our method achieves distribution calibration and robust classification via cross-subspace feature alignment, joint optimization of the global classifier, and confidence-aware knowledge aggregation. Key contributions include: (1) explicit modeling and correction of feature subspace shifts among modular adapters; and (2) dynamic weighting of adapter outputs based on task confidence, significantly improving tolerance to erroneous task ID assignments. Experiments demonstrate that our approach consistently outperforms state-of-the-art PEFT methods across multiple continual learning benchmarks, with particularly notable gains in robustness when task identifiers are corrupted or ambiguous.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• 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

Abstract

Continual Learning (CL) empowers AI models to continuously learn from sequential task streams. Recently, parameter-efficient fine-tuning (PEFT)-based CL methods have garnered increasing attention due to their superior performance. They typically allocate a unique sub-module for learning each task, with a task recognizer to select the appropriate sub-modules for testing images. However, due to the feature subspace misalignment from independently trained sub-modules, these methods tend to produce ambiguous decisions under misleading task-ids. To address this, we propose Cross-subspace Knowledge Alignment and Aggregation (CKAA), a novel framework that enhances model robustness against misleading task-ids through two key innovations: (1) Dual-level Knowledge Alignment (DKA): By aligning intra-class feature distributions across different subspaces and learning a robust global classifier through a feature simulation process, DKA enables the model to distinguish features from both correct and incorrect subspaces during training. (2) Task-Confidence-guided Mixture of Adapters (TC-MoA): A robust inference scheme that adaptively aggregates task-specific knowledge from relevant sub-modules based on task-confidence scores, avoiding overconfidence in misleading task-id predictions. Extensive experiments demonstrate that CKAA outperforms existing PEFT-based CL methods.

Problem

• Aligns feature subspaces to reduce ambiguous decisions
• Enhances robustness against misleading task identifiers
• Aggregates task-specific knowledge adaptively during inference

Innovation

• Aligns intra-class features across subspaces
• Learns robust global classifier via simulation
• Adaptively aggregates knowledge using task-confidence