Research map / Label-efficient and robust learning / OOD and anomaly detection
GEM: Gaussian Embedding Modeling for Out-of-Distribution Detection in GUI Agents
Zheng Wu, Pengzhou Cheng, Zongru Wu, Lingzhong Dong, Zhuosheng Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| fit GMM over embedding distances (GEM) instead of Youden-index thresholding on embedding or uncertainty scores | Qwen2-VL-7B GUI agent · AndroidControl OOD detection | accuracy: 78.20 -> 99.39 | Improved |
| use GEM instead of best-layer embedding on Omniact-Desktop | Qwen2-VL-7B GUI agent · Omniact-Desktop OOD detection | accuracy: 91.15 -> 89.53 accuracy; 80.94 -> 93.55 F1 | Mixed |
| use TV score for OOD detection | GUI agent · Pilot study | AUROC: 54.26 on smartphone | Hurt |
| use uncertainty-based methods (top-k confidence, output entropy) | GUI agent · Pilot study | AUROC: 67.07 smartphone, 57.51 computer | Hurt |
| use encoder from Qwen2/2.5-VL or UI-TARS instead of LLaVA1.5, BLIP, OS-Atlas (ablation) | GEM · Smartphone OOD detection | accuracy: 28.83 (LLaVA1.5) -> 99.78 (Qwen2.5-VL-7B) | Improved |
| route OOD samples to GPT-4o using GEM | SFT Qwen2-VL-7B agent · AITZ + OS-Kairos mixed test | step-wise total accuracy: 49.38 -> 58.78 | Improved |
| use embeddings from intermediate layer (ninth) rather than last layer (ablation) | embedding-based OOD detection, Qwen2-VL-7B · 8 OOD datasets | AUROC | Improved |
| add GEM to training and inference | GUI agent · Time cost | additional time: training +4.9%, testing +6.5% | No clear effect |
Closest papers by meaning
- OS-ATLAS: Foundation Action Model for Generalist GUI Agents (ICLR 2025 · similarity 0.85)
- Faithful Mobile GUI Agents with Guided Advantage Estimator (ICML 2026 · similarity 0.85)
- TongUI: Internet-Scale Trajectories from Multimodal Web Tutorials for Generalized GUI Agents (AAAI 2026 · similarity 0.85)
- MMBench-GUI: A Unified Hierarchical Evaluation Framework for Multi-Platform GUI Agents (CVPR 2026 · similarity 0.85)
- Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents (ICLR 2025 · similarity 0.85)
- D-GARA: A Dynamic Benchmarking Framework for GUI Agent Robustness in Real-World Anomalies (AAAI 2026 · similarity 0.84)
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.
