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..
AMBER: Adaptive Mesh Generation by Iterative Mesh Resolution Prediction
Authors: Niklas Freymuth, Tobias Würth, Nicolas Schreiber, Balázs Gyenes, Andreas Boltres, Johannes Mitsch, Aleksandar Taranovic, Tai Hoang, Philipp Dahlinger, Philipp Becker, Luise Kärger, Gerhard Neumann
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate AMBER on 2D and 3D datasets, including classical physics problems, mechanical components, and real-world industrial designs with human expert meshes. AMBER generalizes to unseen geometries and consistently outperforms multiple recent baselines, including ones using Graph and Convolutional Neural Networks, and Reinforcement Learning-based approaches. |
| Researcher Affiliation | Collaboration | Niklas Freymuth1 Tobias Würth2 Nicolas Schreiber1 Balazs Gyenes1 Andreas Boltres1 3 Johannes Mitsch2 Aleksandar Taranovic1 Tai Hoang1 Philipp Dahlinger1 Philipp Becker1 Luise Kärger2 Gerhard Neumann1 1Autonomous Learning Robots, Karlsruhe Institute of Technology, Karlsruhe 2Institute of Vehicle System Technology, Karlsruhe Institute of Technology, Karlsruhe 3SAP SE |
| Pseudocode | No | The paper describes the methodology using textual explanations, mathematical equations, and diagrams (e.g., Figure 1 for the training pipeline). However, it does not include any explicitly labeled 'Pseudocode', 'Algorithm', or structured code-like blocks. |
| Open Source Code | Yes | Project page, code and datasets are available at https://niklasfreymuth.github.io/AMBER. |
| Open Datasets | Yes | We introduce six novel datasets representing realistic FEM problems that need adaptive meshing to meet common efficiency and accuracy requirements. ... Project page, code and datasets are available at https://niklasfreymuth.github.io/AMBER. |
| Dataset Splits | Yes | Table 3: Number of data points per split and min/mean/max number of vertices and elements per mesh in the training data. ... # Data Points # Vertices # Elements Name Train Val Test Min Mean Max Min Mean Max Poisson (easy) 20 20 20 ... |
| Hardware Specification | Yes | All graph-based methods are trained on an NVIDIA 3090 GPU. The image-based methods are instead trained on an NVIDIA A100 GPU to accommodate the memory requirement of the comparatively high-resolution images. |
| Software Dependencies | No | We implement all neural networks in Py Torch [86] and optimize using ADAM [87]. ... We implement the FEM Poisson and Laplace in SCIKIT-FEM [76]. ... We use GMSH [25] for mesh generation. The paper mentions various software components but does not provide specific version numbers for them. |
| Experiment Setup | Yes | Table 4: AMBER parameters and experiment configuration (variable names as used in the main text) Section Parameter Variable Value Optimization Optimizer ADAM Learning rate 1.0 10 3 Learning rate scheduler linear with 10 % warm-up Weight decay 1.0 10 6 Aggregation function L mean MPN steps L 20 Activation function Leaky Re LU Edge dropout 0.1 MLP layers 2 Latent dimension 64 Refinement steps T 3 Maximum buffer size 500 meshes Buffer addition frequency k 8 samples every 128 batches Training steps 25 600 or 51 200 (task-dependent) Batch size 500 000 graph nodes plus edges Sizing field scaling ct 1.618T t 1 |