Research map / Probabilistic and scientific modeling / Molecules and proteins
AutoMat: Physics-Guided Agentic Reasoning for Solving Ill-Posed Inverse Microscopy Problems
Yaotian Yang, Yiwen Tang, Yizhe Chen, Xiao Chen, Jiangjie Qiu, Hao Xiong, Haoyu Yin, Zhiyao Luo, Yifei Zhang, Sijia Tao, Wentao Li, Qinghua Zhang and 5 more
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 |
|---|---|---|---|
| use the LLM failure-aware controller instead of a hand-coded heuristic script with same tools (ablation) | AutoMat · STEM2Mat-Bench overall | formation energy MAE (meV/atom): 385.42 -> 321.57 | Improved |
| use the LLM controller instead of heuristic script on hardest tier (ablation) | AutoMat · STEM2Mat-Bench Tier 3 | formation energy MAE (meV/atom): 585.00 -> 333.49 | Improved |
| use the LLM controller instead of heuristic script (ablation) | AutoMat · STEM2Mat-Bench Tier 3 | structure success rate S.S. (%): 62.3 -> 73.1 | Improved |
| use the LLM controller instead of heuristic script (ablation) | AutoMat · STEM2Mat-Bench overall | composition correctness C.C. (%): 76.5 -> 83.2 | Improved |
| use the LLM controller instead of heuristic script (ablation) | AutoMat · STEM2Mat-Bench overall | structure success rate S.S. (%): 74.2 -> 83.2 | Improved |
| swap DeepSeek backbone for GPT-4o (ablation) | AutoMat · STEM2Mat-Bench avg over tiers | formation energy MAE (meV/atom): 321.57 -> 323.25 | No clear effect |
| swap DeepSeek backbone for GPT-4o (ablation) | AutoMat · STEM2Mat-Bench Tier 2 | in-plane lattice RMSD (A): 0.11 -> 0.12 | No clear effect |
| use AutoMat instead of off-the-shelf MicroscopyGPT | AutoMat · STEM2Mat-Bench avg over tiers | formation energy MAE (meV/atom): 620.22 -> 321.57 | Improved |
| use AutoMat instead of MicroscopyGPT | AutoMat · STEM2Mat-Bench Tier 2 | structure success rate S.S. (%): 25.9 -> 84.0 | Improved |
| use AutoMat instead of general VLMs such as GPT-4.1 Vision | AutoMat · STEM2Mat-Bench avg over tiers | formation energy MAE (meV/atom): 2657.3 -> 321.57 | Improved |
Closest papers by meaning
- AutoMS: Multi-Agent Evolutionary Search for Cross-Physics Inverse Microstructure Design (ICML 2026 · similarity 0.87)
- Co-PiLOT: Constrained Physics-Informed Latent Optimization for Target-Driven Inverse Design (NeurIPS 2026 · similarity 0.86)
- EP-Flow: Disordered Crystal Structure Prediction without Site-level Annotation (NeurIPS 2026 · similarity 0.85)
- LLM Meets Diffusion: A Hybrid Framework for Crystal Material Generation (NeurIPS 2025 · similarity 0.85)
- Scalable Diffusion for Materials Generation (ICLR 2024 · similarity 0.85)
- All-atom Diffusion Transformers: Unified generative modelling of molecules and materials (ICML 2025 · similarity 0.85)
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.
