From b912fe0e74c098386560a1aaba7a7ced98978afc Mon Sep 17 00:00:00 2001 From: Bochengz Date: Thu, 17 Apr 2025 11:50:06 +0800 Subject: [PATCH] update phympgn README --- .../applications/data_mechanism_fusion/phympgn/README.md | 9 ++++++--- .../data_mechanism_fusion/phympgn/README_CN.md | 9 ++++++--- 2 files changed, 12 insertions(+), 6 deletions(-) diff --git a/MindFlow/applications/data_mechanism_fusion/phympgn/README.md b/MindFlow/applications/data_mechanism_fusion/phympgn/README.md index 5dfb7543f..057b98159 100644 --- a/MindFlow/applications/data_mechanism_fusion/phympgn/README.md +++ b/MindFlow/applications/data_mechanism_fusion/phympgn/README.md @@ -10,7 +10,9 @@ Physics-encoded Message Passing Graph Network (PhyMPGN) is capable to model spat - Considering the universality of diffusion processes in physical phenomena, a learnable Laplace Block is designed, which encodes the discrete Laplace-Beltrami operator - A novel padding strategy to encode different types of BCs into the learning model is proposed. -Paper link: [https://arxiv.org/abs/2410.01337](https://gitee.com/link?target=https%3A%2F%2Farxiv.org%2Fabs%2F2410.01337) +Paper link: [https://arxiv.org/abs/2410.01337](https://gitee.com/link?target=https%3A%2F%2Farxiv.org%2Fabs%2F2410.01337). + +This paper has been accepted as ICLR 2025 Spotlight, see https://openreview.net/forum?id=fU8H4lzkIm¬eId=wS5SaVKjWt. ## Problem Setup @@ -107,7 +109,8 @@ $Re=480$ | Dataset | Cylinder flow | | Model Parameters | 950k | | Training Configuration | batch_size=4,
epochs=1600 | -| Training Loss
(MSE) | | -| Inference Loss
(MSE) | | +| Training Loss
(MSE) | 3.05e-5 | +| Validation Loss
(MSE) | 5.58e-6 | +| Inference MSE | 4.88e-2 | | Training Speed
(s / epoch) | 420 s | | Inference Speed
(s / trajectory) | 174 s | diff --git a/MindFlow/applications/data_mechanism_fusion/phympgn/README_CN.md b/MindFlow/applications/data_mechanism_fusion/phympgn/README_CN.md index 55ae876f7..8b19fe6d5 100644 --- a/MindFlow/applications/data_mechanism_fusion/phympgn/README_CN.md +++ b/MindFlow/applications/data_mechanism_fusion/phympgn/README_CN.md @@ -8,7 +8,9 @@ - 考虑到物理现象中普遍存在扩散过程，设计了一个可学习的Laplace Block，编码了离散拉普拉斯-贝尔特拉米算子（Laplace-Beltrami Operator） - 提出了一个新颖的填充策略在模型中编码不同类型的边界条件 -论文链接: [https://arxiv.org/abs/2410.01337](https://gitee.com/link?target=https%3A%2F%2Farxiv.org%2Fabs%2F2410.01337) +论文链接: [https://arxiv.org/abs/2410.01337](https://gitee.com/link?target=https%3A%2F%2Farxiv.org%2Fabs%2F2410.01337)。 + +该论文被接收为 ICLR 2025 Spotlight，详见 https://openreview.net/forum?id=fU8H4lzkIm¬eId=wS5SaVKjWt。 ## 问题描述 @@ -105,7 +107,8 @@ $Re=480$ | 数据集 | Cylinder flow | | 参数量 | 950k | | 训练参数 | batch_size=4,
epochs=1600 | -| 训练损失
(MSE) | | -| 推理损失
(MSE) | | +| 训练损失
(MSE) | 3.05e-5 | +| 验证损失
(MSE) | 5.58e-6 | +| 推理误差
(MSE) | 4.88e-2 | | 训练速度
(s / epoch) | 420 s | | 推理速度
(s / trajectory) | 174 s | -- Gitee