Learning Symbolic Models for Graph-structured Physical Mechanism
Authors: Hongzhi Shi, Jingtao Ding, Yufan Cao, quanming yao, Li Liu, Yong Li
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on datasets from different physical domains, including mechanics, electricity, and thermology, and on real-world datasets of pedestrian dynamics without ground-truth formulas. The experimental results not only verify the rationale of our design but also demonstrate that the proposed method can automatically learn precise and interpretable formulas for graph-structured physical mechanisms. |
| Researcher Affiliation | Collaboration | Hongzhi Shi1, Jingtao Ding1, Yufan Cao1, Quanming Yao1, Li Liu2, Yong Li1. Department of Electronic Engineering, Tsinghua University1. Biren Tech2. |
| Pseudocode | Yes | Algorithm 1 The process of obtaining the graph-structured symbolic model. |
| Open Source Code | No | The paper mentions using a third-party tool, PySR, and provides its GitHub link: '4https://github.com/Miles Cranmer/Py SR'. However, it does not provide an explicit statement or link for the source code of the methodology described in this paper. |
| Open Datasets | Yes | We conduct experiments on two real-world datasets of crowd trajectories: several experimental datasets from studies about pedestrian dynamics (Boltes & Seyfried, 2013)1, including the following scenarios, (i) Unidirectional flow in corridor: a group of people goes through a corridor in the same direction, as shown in Figure 8(a); (ii) Bidirectional flow in corridor: a group of people goes through a corridor in opposite directions, as shown in Figure 8(b). 1https://ped.fz-juelich.de/database/doku.php |
| Dataset Splits | No | The paper mentions '# X-Y pairs' for each dataset in Table 3 and refers to 'training data' in Algorithm 1, but it does not provide specific train/validation/test dataset split percentages, absolute sample counts, or explicit instructions for how to create these splits. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory, cloud instances) used to run its experiments. |
| Software Dependencies | No | The paper states: 'We implement our model in Python using Pytorch library and optimize all the models by Adam optimizer (Kingma & Ba, 2015). We use parallel symbolic regression in Python (Py SR)4 (Cranmer, 2020) to extract formulas from each message functions ϕ.' However, it does not provide specific version numbers for Python, PyTorch, or PySR. |
| Experiment Setup | Yes | For the DL part, we set the learning rate as 10 4, tolerance in early stopping as 10, #layers and embedding size in MLP as 4 and 16, the max number of epochs as 20000 and the weight λ as 0.1... For the traditional SR part, our candidate symbols include both binary operator {+, , , /} and unary operator {sign, exp}, and we set batch size as 1024. |