AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

CGL: Advancing Continual GUI Learning via Reinforcement Fine-Tuning

Zhenquan Yao, Zitong Huang, Yihan Zeng, Jianhua Han, Hang Xu, Chun-Mei Feng, Jianwei Ma, Wangmeng Zuo

CVPR 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Mixture-of-experts for multimodal continual tuning · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add GRPO with KL instead of SFT only (ablation)Qwen2.5-VL-3B · AndroidControl-CL Order 1avg step accuracy / FM: 76.90 -> 81.53; FM -5.73 -> -0.62Improved
add KL constraint to SFT (ablation)SFT · AndroidControl-CL Order 1avg step accuracy: 76.90 -> 80.84Improved
remove KL from GRPO (ablation)GRPO · AndroidControl-CL Order 1avg step accuracy: 81.53 -> collapse (unavailable)Hurt
add 5% historical replay to SFT (ablation)SFT · AndroidControl-CL Order 1avg step accuracy: 76.90 -> 79.80Improved
add static full SFT to GRPO (ablation)GRPO+KL · AndroidControl-CL Order 1avg step accuracy: 81.53 -> 81.68Improved
replace static SFT with error-aware dynamic SFT (ablation)SFT+GRPO · AndroidControl-CL Order 1avg step accuracy: 81.68 -> 81.90Improved
add entropy-regulated lambda (ablation)D-SFT+GRPO · AndroidControl-CL Order 1avg step accuracy: 81.90 -> 82.23Improved
add gradient surgery (ablation)D-SFT+GRPO · AndroidControl-CL Order 1avg step accuracy / FM: 81.90 -> 82.10; FM -0.33 -> -0.05Improved
use dynamic lambda instead of fixed lambda 1 (ablation)CGL · AndroidControl-CL Order 1avg step accuracy: 81.62 -> 82.33Improved
use dynamic lambda instead of fixed lambda 0.01 (ablation)CGL · AndroidControl-CL Order 1avg step accuracy: 81.90 -> 82.33Improved
CGL vs GRPO across three task ordersQwen2.5-VL-3B · AndroidControl-CL Orders 1-3avg step accuracy: GRPO 81.44-81.75 -> CGL 82.33-82.61Improved
CGL vs SFT forgetting on LLaVA-OV-0.5BLLaVA-OneVision-0.5B · AndroidControl-CL Order 1FM: -6.81 -> -0.52Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy