DiffPDE: Diffusion Models for PDE Solvers
DiffPDE: Masked Diffusion Language Models as PDE Solver
DiffPDE offers faster and more efficient PDE solver repair than traditional autoregressive models, potentially reducing computational costs for researchers and engineers in ASEAN's growing tech sector.
DiffPDE uses discrete diffusion language models for targeted code repair in Partial Differential Equation solvers.
The framework introduces localized re-masking and infilling to regenerate only erroneous regions while preserving correct context.
On PDEBench, DiffPDE achieves competitive accuracy and significantly accelerates repair compared to same-scale autoregressive models.
The approach also includes Iterative Debugging GRPO for handling coupled bugs requiring sequential interventions.
DiffPDE: Masked Diffusion Language Models as PDE Solver
Existing approaches for synthesizing Partial Differential Equation (PDE) solvers predominantly rely on autoregressive models, yet their global left-to-right decoding incurs substantial redundancy when addressing inherently localized bugs. In this work, we challenge this inefficient paradigm and propose DiffPDE, a framework leveraging discrete diffusion language models for targeted code repair. By introducing a localized re-masking and infilling strategy, DiffPDE regenerates only erroneous regions while preserving correct context, naturally aligning generation with the sparse nature of PDE errors. Furthermore, to handle coupled bugs requiring sequential interventions, we present Iterative Debugging GRPO (ID-GRPO), a reinforcement learning scheme that enables multi-round debugging within single trajectories via intermediate rewards. Experiments on PDEBench show that DiffPDE achieves competitive accuracy, outperforms same-scale AR models, and significantly accelerates repair.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include DiffPDE, PDE, diffusion models, code repair.
Source attribution: curated from arXiv cs.AI (@Wenxuan Guo, Yuyang Hong, Lubin Fan, Zhaojin Fu, Lin Chen, Kun Ding, Shiming Xiang).
