Split-on-Share: Mixture of Sparse Experts for Task-Agnostic Continual Learning

TL;DR

SETA introduces a modular mixture of sparse experts to improve task-agnostic continual learning in LLMs, effectively balancing plasticity and stability by isolating task-specific and shared knowledge.

Contribution

It proposes a novel framework that decomposes models into experts and shared components, using elastic weight anchoring and gating for improved continual learning performance.

Findings

SETA outperforms existing parameter-efficient methods across multiple benchmarks.

The expert-shared decomposition effectively mitigates catastrophic forgetting.

Elastic weight anchoring preserves critical shared knowledge during updates.

Abstract

Continual learning in Large Language Models (LLMs) is hindered by the plasticity-stability dilemma, where acquiring new capabilities often leads to catastrophic forgetting of previous knowledge. Existing methods typically treat parameters uniformly, failing to distinguish between specific task knowledge and shared capabilities. We introduce Mixture of Sparse Experts for Task-Agnostic Continual Learning, referred to as SETA, a framework that resolves the plasticity-stability conflict by decomposing the model into modular subspaces. Unlike standard updates, where tasks compete for the same parameters, SETA separates knowledge into unique experts, designed to isolate task-specific patterns, and shared experts, responsible for capturing common features. This structure is maintained through elastic weight anchoring, which protects critical shared knowledge and enables a unified gating network to automatically retrieve the correct expert combination for each task during inference. Extensive experiments across diverse domain-specific and general benchmarks demonstrate that SETA consistently outperforms state-of-the-art parameter-efficient fine-tuning-based continual learning methods.