PASs-MoE: Mitigating Misaligned Co-drift among Router and Experts via Pathway Activation Subspaces for Continual Learning

TL;DR

This paper introduces PASs-MoE, a method that uses pathway activation subspaces to align routing and expert adaptation in continual learning, reducing forgetting and improving performance without extra parameters.

Contribution

It proposes a novel PASs-based MoE-LoRA approach that mitigates misaligned co-drift by calibrating routing and stabilizing important rank directions.

Findings

Outperforms conventional continual learning baselines.

Reduces forgetting in multimodal large language models.

Improves accuracy on continual instruction tuning benchmarks.

Abstract

Continual instruction tuning (CIT) requires multimodal large language models (MLLMs) to adapt to a stream of tasks without forgetting prior capabilities. A common strategy is to isolate updates by routing inputs to different LoRA experts. However, existing LoRA-based Mixture-of-Experts (MoE) methods often jointly update the router and experts in an indiscriminate way, causing the router's preferences to co-drift with experts' adaptation pathways and gradually deviate from early-stage input-expert specialization. We term this phenomenon Misaligned Co-drift, which blurs expert responsibilities and exacerbates forgetting.To address this, we introduce the pathway activation subspace (PASs), a LoRA-induced subspace that reflects which low-rank pathway directions an input activates in each expert, providing a capability-aligned coordinate system for routing and preservation. Based on PASs, we propose a fixed-capacity PASs-based MoE-LoRA method with two components: PAS-guided Reweighting, which calibrates routing using each expert's pathway activation signals, and PAS-aware Rank Stabilization, which selectively stabilizes rank directions important to previous tasks. Experiments on a CIT benchmark show that our approach consistently outperforms a range of conventional continual learning baselines and MoE-LoRA variants in both accuracy and anti-forgetting without adding parameters. Our code will be released upon acceptance.