BPG: Balancing Plasticity and Generalization for Domain Incremental Learning

2026-08-11

AI Summary

This work addresses the vulnerability of deep neural networks to distribution shifts in domain-incremental learning, where existing approaches often suffer from insufficient learning capacity or excessive parameter redundancy. To overcome these limitations, the authors propose BPG, a unified framework that innovatively integrates dynamic adapter dimension allocation based on feature separability (BPG-Adapter) with a soft multi-model fusion inference mechanism (BPG-Inference). This design maintains parameter isolation while simultaneously enhancing model plasticity and generalization. Extensive experiments demonstrate that BPG achieves state-of-the-art average accuracy on DomainNet, CDDB, and CORe50 benchmarks, with a remarkably low forgetting rate of 0.22% on DomainNet—significantly outperforming baselines employing uniform adapters or hard domain selection strategies.

Abstract

Deep neural networks excel in various tasks but struggle to generalize across evolving data distributions, leading to significant performance degradation under domain shifts. Domain incremental learning (DIL) addresses this challenge by enabling models to continuously adapt while retaining prior knowledge. Among existing DIL approaches, the parameter-isolation paradigm achieves state-of-the-art performance. However, these methods often adopt a one-size-fits-all approach to adapt to new domains, resulting in either insufficient learning capacity or redundant parameters. In this work, we propose BPG, a unified framework that addresses both challenges through two complementary components: BPG-Adapter, which dynamically determines each domain's adapter hidden dimension based on domain-specific feature separability, and BPG-Inference, a soft domain mixture strategy that integrates multiple domain-specific models at test time, mitigating domain ID misselection. Experimental results on DomainNet, CDDB, and CORe50 demonstrate that BPG consistently outperforms uniform adapter-based approaches and hard domain selection strategies, achieving state-of-the-art average accuracy while reducing forgetting to as low as 0.22% on DomainNet.

Problem

• domain incremental learning
• plasticity
• generalization
• domain shift
• parameter redundancy

Innovation

• domain incremental learning
• parameter isolation
• dynamic adapter
• soft domain mixture
• feature separability