Research map / Reinforcement learning / RL from feedback
LithoGRPO: Fast Inverse Lithography via GRPO Reinforced Flow Matching
Yao Lai, Xuyuan Xiong, Zeyue Xue, Guojin Chen, Jing Wang, Xihui Liu, Rui Zhang, Robert D. Mullins, Bei Yu, Ping Luo
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 |
|---|---|---|---|
| add GRPO RL fine-tuning after SFT (ablation) | LithoGRPO · MetalSet | Shot count: 803 -> 444 | Improved |
| add GRPO RL fine-tuning after SFT (ablation) | LithoGRPO · ViaSet | Shot count: 989 -> 398 | Improved |
| add GRPO RL fine-tuning after SFT (ablation) | LithoGRPO · MetalSet | PVB: 40722 -> 39838 | Improved |
| add GRPO RL fine-tuning after SFT (ablation) | LithoGRPO · MetalSet | L2: 29123 -> 28933 | Improved |
| add GRPO RL fine-tuning after SFT (ablation) | LithoGRPO · ILT average over 4 datasets | average rank: 4.7 -> 4.3 | Improved |
| add SFT on differentiable metrics after pretraining (ablation) | LithoGRPO · MetalSet | L2: 32824 -> 29123 | Improved |
| add SFT on differentiable metrics after pretraining (ablation) | LithoGRPO · MetalSet | Shot count: 487 -> 803 | Hurt |
| use two sampling steps instead of one (ablation) | LithoGRPO · Metal layers | average rank: 4.7 -> 3.8 | Improved |
| use ten sampling steps instead of one (ablation) | LithoGRPO · Metal layers | average rank: 4.7 -> 6.7 | Hurt |
| use two sampling steps instead of one (ablation) | LithoGRPO · Metal layers | runtime (s): 0.104 -> 0.203 | Hurt |
| use colored noise instead of Gaussian noise in SDE exploration (ablation) | LithoGRPO · MetalSet | shot count | Improved |
| lower noise level from a=0.1 to a=0.01 (ablation) | LithoGRPO · MetalSet | reward convergence speed | Hurt |
Closest papers by meaning
- TokMan:Tokenize Manhattan Mask Optimization for Inverse Lithography (NeurIPS 2025 · similarity 0.90)
- LithoDreamer: A Physics-Informed World Model for Multi-Stage Computational Lithography (ICML 2026 · similarity 0.90)
- ILILT: Implicit Learning of Inverse Lithography Technologies (ICML 2024 · similarity 0.90)
- Diff-Instruct with Diffused Reward: Towards Principled One-step Generator RL (NeurIPS 2026 · similarity 0.83)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.82)
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.81)
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.
