SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2509.20570 · NeurIPS · 2025

PIRF: Physics-Informed Reward Fine-Tuning for Diffusion Models

Na Li, Pengfei Jin, Quanzheng Li, Mingze Yuan

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 10 functions out of this paper's own repositories and ran 4 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
mingze-yuan/PIRF canonical 4 of 10
FunctionStatusWhere it lives
format_time Ran mingze-yuan/PIRF/dnnlib/util.py
code served (permissive licence) · get_code("674eca7b9e1b6439")
format_time_brief Ran mingze-yuan/PIRF/dnnlib/util.py
code served (permissive licence) · get_code("32af001972a0699f")
parse_int_list Ran mingze-yuan/PIRF/finetune_pirf.py
code served (permissive licence) · get_code("cacd4f6ec202d9b4")
weight_init Ran mingze-yuan/PIRF/training/networks.py
code served (permissive licence) · get_code("d41a4250066bce93")
ask_yes_no Not yet run mingze-yuan/PIRF/dnnlib/util.py
code served (permissive licence) · get_code("9d31d2c4cd16bb2d")
generate Not yet run mingze-yuan/PIRF/evaluate_pirf.py
code served (permissive licence) · get_code("3dcb7d7bb60bc495")
generate_with_cocogen Not yet run mingze-yuan/PIRF/baselines/evaluate_cocogen.py
code served (permissive licence) · get_code("575ce4b10b86c029")
generate_with_diffusionpde Not yet run mingze-yuan/PIRF/baselines/evaluate_diffusionpde.py
code served (permissive licence) · get_code("3a630d8b0e06c2b6")
generate_with_pgdiffusion Not yet run mingze-yuan/PIRF/baselines/evaluate_pgdiffusion.py
code served (permissive licence) · get_code("b2b6f79b379737a5")
generate_with_pidm Not yet run mingze-yuan/PIRF/baselines/evaluate_pidm.py
code served (permissive licence) · get_code("29ba7759e852cc27")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Diffusion models have demonstrated strong generative capabilities across scientific domains, but often produce outputs that violate physical laws. We propose a new perspective by framing physics-informed generation as a sparse reward optimization problem, where adherence to physical constraints is treated as a reward signal. This formulation unifies prior approaches under a reward-based paradigm and reveals a shared bottleneck: reliance on diffusion posterior sampling (DPS)-style value function approximations, which introduce non-negligible errors and lead to training instability and inference inefficiency. To overcome this, we introduce Physics-Informed Reward Fine-tuning (PIRF) -a method that bypasses value approximation by computing trajectory-level rewards and backpropagating their gradients directly. However, a naive implementation suffers from low sample efficiency and compromised data fidelity. PIRF mitigates these issues through two key strategies: (1) a layer-wise truncated backpropagation method that leverages the spatiotemporally localized nature of physics-based rewards, and (2) a weight-based regularization scheme that improves efficiency over traditional distillation-based methods. Across five PDE benchmarks, PIRF consistently achieves superior physical enforcement under efficient sampling regimes, highlighting the potential of reward fine-tuning for advancing scientific generative modeling. Our code is available at https://github.com/mingze-yuan/PIRF.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2509.20570")
get_code_for_paper("2509.20570")
have("2509.20570")

Connect an agent — have() is free.