Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
GUI-Reflection: Empowering Multimodal GUI Models with Self-Reflection Behavior
Authors: Penghao Wu, Shengnan Ma, Bo Wang, Jiaheng Yu, Lewei Lu, Ziwei Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 3 Experiments In this part, we continue from the GUIPretrain-Ref model and conduct experiments to validate the effectiveness of reflection data in the SFT and online stages. First, we verify the effect of augmenting offline GUI SFT data with reflection data and conducting iterative reflection tuning in the online environment. We conduct experiments in our GUI environment by training models with the level-1 tasks for 3 iterations and evaluating the performance on the level-2 tasks. As shown in Table 2, the baseline model trained without reflection data in offline SFT and using only filtered BC achieves a success rate of 14.58% on level-2 tasks. |
| Researcher Affiliation | Collaboration | Penghao Wu , Shengnan Ma , Bo Wang , Jiaheng Yu , Lewei Lu , Ziwei Liu S-Lab, Nanyang Technological University , Sense Time Research Project Page: https://penghao-wu.github.io/GUI_Reflection/ Corresponding authors: EMAIL |
| Pseudocode | No | The paper describes the 'Iterative Online Reflection Tuning algorithm' in Section 2.5.2 and illustrates it in Figure 4, but it does not present the algorithm in a structured pseudocode block or a clearly labeled algorithm section with code-like formatting. |
| Open Source Code | No | The abstract states: 'with all data, models, environments, and tools to be released publicly.' However, the NeurIPS Paper Checklist for Question 5 explicitly states: 'Answer: [No] Justification: We do not provide access to the code or data during the submission phase.' |
| Open Datasets | Yes | For the Action Verification and Action Reversal tasks, the training and evaluation data are constructed from the training and test splits of Android Control [21] and GUI-Odyssey [27]. For the GUI offline SFT stage, we use public mobile device GUI interaction datasets including AITW [33], AITZ [50], AMEX [11], GUI-Odyssey [27], and Android Control [21]. |
| Dataset Splits | Yes | For the Action Verification and Action Reversal tasks, the training and evaluation data are constructed from the training and test splits of Android Control [21] and GUI-Odyssey [27]. We have 1206 and 420 samples for the evaluation of these two tasks, respectively... For the Mistake-informed reattempt task, the training data is constructed from Wave-UI [2], AMEX [11], and OS-ATLAS-Desktop [43]. For this task, we evaluate directly on Screen Spot [14] and Screen Spotv2 [43]. |
| Hardware Specification | Yes | All the training is conducted on 32 H100 GPUs. |
| Software Dependencies | No | The paper mentions using 'Adam W [26] optimizer' but does not specify any programming languages, libraries, or frameworks with their version numbers that were used for implementation or training. |
| Experiment Setup | Yes | For the GUI pre-training, we train the model for 1 epoch with a learning rate of 4 10 5. For the SFT stage, we train the pre-trained model for 1 epoch with a learning rate of 3 10 5. In each reflection tuning iteration, we train the model on the collected data in this iteration for 2 epochs with a learning rate of 1 10 5. We use Adam W [26] optimizer for all the training. |