Adaptive Model Ensemble for Continual Learning

2025-09-24

AI Summary

To address task-interference and intra-layer knowledge conflicts in model ensembling—key contributors to catastrophic forgetting in continual learning—this paper proposes a meta-learning-driven adaptive hierarchical ensembling method. Our approach introduces a lightweight mixture coefficient generator that takes task identifiers as input and dynamically produces layer-wise parameter interpolation weights, enabling task- and layer-aware knowledge fusion without architectural modifications to the backbone network. The method is plug-and-play, readily enhancing mainstream continual learning algorithms. Extensive experiments on benchmarks including Split-CIFAR100 and ImageNet-R demonstrate substantial mitigation of forgetting: our method achieves average accuracy improvements of 2.3–4.7 percentage points over state-of-the-art approaches, validating its effectiveness and generalizability across diverse continual learning scenarios.

Technology Category

• Machine Learning: Ensemble Methods
• Search and Optimization: Learning to Search
• 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

Model ensemble is an effective strategy in continual learning, which alleviates catastrophic forgetting by interpolating model parameters, achieving knowledge fusion learned from different tasks. However, existing model ensemble methods usually encounter the knowledge conflict issue at task and layer levels, causing compromised learning performance in both old and new tasks. To solve this issue, we propose meta-weight-ensembler that adaptively fuses knowledge of different tasks for continual learning. Concretely, we employ a mixing coefficient generator trained via meta-learning to generate appropriate mixing coefficients for model ensemble to address the task-level knowledge conflict. The mixing coefficient is individually generated for each layer to address the layer-level knowledge conflict. In this way, we learn the prior knowledge about adaptively accumulating knowledge of different tasks in a fused model, achieving efficient learning in both old and new tasks. Meta-weight-ensembler can be flexibly combined with existing continual learning methods to boost their ability of alleviating catastrophic forgetting. Experiments on multiple continual learning datasets show that meta-weight-ensembler effectively alleviates catastrophic forgetting and achieves state-of-the-art performance.

Problem

• Addresses knowledge conflicts in model ensemble methods for continual learning
• Solves task-level and layer-level knowledge conflicts using adaptive fusion
• Alleviates catastrophic forgetting while learning new tasks efficiently

Innovation

• Meta-learning trains generator for adaptive model ensemble coefficients
• Layer-specific coefficients resolve task and layer knowledge conflicts
• Flexible integration with existing methods boosts forgetting prevention