Work

Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning

Year: 2026

Type: conference-paper

Abstract: Large Language Model (LLM)-based agents significantly extend the utility of LLMs by interacting with dynamic environments.However, enabling agents to continually learn new tasks without catastrophic forgetting remains a critical challenge, known as the stability-plasticity dilemma.In this work, we argue that this dilemma fundamentally arises from the failure to explicitly distinguish between common knowledge shared across tasks and conflicting knowledge introduced by task-specific interference.To address this, we propose Agent-Dice, a parameter fusion framework based on directional consensus evaluation.Concretely, Agent-Dice disentangles knowledge updates through a twostage process: geometric consensus filtering to prune conflicting gradients, and curvaturebased importance weighting to amplify shared semantics.We provide a rigorous theoretical analysis that establishes the validity of the proposed fusion scheme and offers insight into the origins of the stability-plasticity dilemma.Extensive experiments on GUI agents and tooluse agent domains demonstrate that Agent-Dice exhibits outstanding continual learning performance with minimal computational overhead and parameter updates.The codes are available at https://github.com/

Source: Findings of the Association for Computational Linguistics: ACL (Findings of ACL)

DOI: 10.18653/v1/2026.findings-acl.908