GUI-CIDER: Mid-training GUI Agents via Causal Internalization and Density-aware Exemplar Reselection

arXiv:2605.28534 · 2026-05-28 공개 · arXiv · PDF

reinforcement-learning supervised-fine-tuning gui-agents task-completion mid-training trajectory-distillation causal-internalization exemplar-reselection

Abstract

Despite the rapid progress of multimodal large language models in building Graphical User Interface (GUI) agents, their real-world task completion is fundamentally bottlenecked by a lack of world knowledge about GUI operations. Existing solutions typically rely on expensive multi-agent scaffolding or conventional post-training paradigms, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). However, post-training only allows agents to implicitly absorb world knowledge through action annotations or reward signals, leading to inefficient trajectory memorization rather than genuine comprehension. Therefore, an approach that enables explicit learning of this knowledge is imperative. To this end, we propose GUI-CIDER, a mid-training method that explicitly internalizes GUI world knowledge through Causal Internalization and Density-aware Exemplar Reselection. GUI-CIDER operates in three stages: (1) data synthesis, which distills static planning and dynamic causal knowledge from GUI trajectories into text; (2) exemplar reselection, which filters the corpus by rewarding causal structures and penalizing semantic redundancy; and (3) mid-training, where the refined data is used to embed the acquired knowledge. Extensive experiments on two GUI knowledge benchmarks and three task completion benchmarks demonstrate that GUI-CIDER consistently improves both the agent's understanding of GUI operations and its task success rates.The codes are available at https://github.com/Wuzheng02/GUI-CIDER.

한국어 요약

📋 한 줄 요약

**[GUI 에이전트 / Mid-Training]** GUI-CIDER가 SFT·RL의 implicit knowledge 흡수 한계를 넘어 mid-training으로 causal·planning 지식을 explicit 내재화 — 3 단계(data synthesis·exemplar reselection·mid-training)로 2 GUI knowledge·3 task completion 벤치 일관 향상.

🎯 핵심 기여도

💡 핵심 아이디어

GUI 에이전트의 진정한 능력은 action annotation을 memorize하는 implicit post-training으로는 도달할 수 없고, trajectory에서 causal·planning 지식을 explicit 텍스트로 distill해 mid-training으로 embed하면 SFT·RL이 따라잡지 못하는 이해 수준에 도달할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: GUI 에이전트 학습 패러다임에 mid-training 단계 정립, causal·planning 지식의 explicit 텍스트 distillation 방법론 제시, 평가 두 축(GUI knowledge·task completion) 모두에서 일관 향상. **한계**: Causal·planning 지식 distillation의 품질·커버리지가 source trajectory 의존, exemplar reselection metric의 hyperparameter 부담, GUI 외 domain(robot·web 외)으로 일반화 미검증.

🚀 실용적 활용