On the Expressive Power of Sparse Geometric MPNNs

Authors: Yonatan Sverdlov, Nadav Dym

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We introduce a simple architecture, EGENNET, which achieves our theoretical guarantees and compares favorably with alternative architectures on synthetic and chemical benchmarks. Our code is available at Git Hub. ... We experimentally find that EGENNET is highly successful in synthetic separation tasks and on several learning tasks involving medium-sized molecules, with a performance comparable to, or better than, state-of-the-art methods. ... 6 EXPERIMENTS In the experiments section, we have three main objectives. First, we aim to demonstrate that invariant models are indeed strictly less expressive than equivariant models and that power graphs help to mitigate this gap. Second, we seek to show that our model can distinguish challenging graph examples, including those requiring multiple layers for separation, thereby avoiding the bottleneck phenomenon encountered in other geometric models. Finally, we aim to demonstrate the improved performance of our model on real-world benchmarks.
Researcher Affiliation Academia Yonatan Sverdlov1, Nadav Dym1,2 1 Faculty of Mathematics 2 Faculty of Computer Science Technion Israel Institute of Technology EMAIL EMAIL
Pseudocode No The paper describes the EGENNET architecture with mathematical equations in Section 5.1 (e.g., equations 4, 5, 6), but it does not include a clearly labeled pseudocode or algorithm block with structured steps.
Open Source Code Yes Our code is available at Git Hub.
Open Datasets Yes The results are shown in the first column in Table 1. As expected, EGENNET perfectly separates pairs a and b, while Sch Net fails to separate both. ... Our task is learning chemical property prediction on three datasets: Drugs (Axelrod & Gomez-Bombarelli, 2022), Kraken (Gensch et al., 2022), and BDE (Meyer et al., 2018).
Dataset Splits Yes For chemical property experiments, all the procedure is taken from Zhu et al. (2024): each dataset is partitioned randomly into three subsets: 70% for training, 10% for validation, and 20% for test.
Hardware Specification Yes All experiments are conducted on servers with NVIDIA A40 GPUs, each with 48GB of memory.
Software Dependencies No We utilize Py Torch and Py Torch-Geometric to implement all deep learning models. While PyTorch and PyTorch-Geometric are mentioned, specific version numbers are not provided.
Experiment Setup Yes For all experiments, we use Adam W optimizer Loshchilov & Hutter (2017). We use a varying weight decay of 1e 4, 1e 5 and learning rate of 1e 4. Reduce On Platue scheduler is used with a rate decrease of 0.75 with the patience of 15 epochs. We use two blocks with 40 channels in the challenging point cloud separation experiments. For k-chain experiments, we repeat each experiment 10 times, with 10 different seeds, and set the number of channels to be 60, and a varying number of blocks, as detailed in the table. For power graph experiments, we repeat each experiment 10 times, with 10 different seeds, and set the number of channels to be 60 and 3 blocks. For chemical property experiments, all the procedure is taken from Zhu et al. (2024): each dataset is partitioned randomly into three subsets: 70% for training, 10% for validation, and 20% for test. We set the number of channels to be 256 and used 6 blocks. Each model is trained over 1,500 epochs. Experiments are repeated three times for all eight regression targets, and the results reported correspond to the model that performs best on the validation set in terms of Mean Absolute Error (MAE).