Research map / Architectures and efficiency / RNNs and normalization
TensorNet: Cartesian Tensor Representations for Efficient Learning of Molecular Potentials
Guillem Simeon, Gianni De Fabritiis
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Cartesian rank-2 tensor representations in TensorNet | molecular potential prediction · QM9 U0 | MAE: 4.1 -> 3.9(1) | Improved |
| use Cartesian rank-2 tensor representations in TensorNet | molecular potential prediction · QM9 G | MAE: 5.5 -> 5.7(1) | Hurt |
| use TensorNet instead of ET | ANI1x training and evaluation · ANI1x | energy MAE: 21.2 -> 17.3 | Improved |
| use TensorNet instead of ET | ANI1x-trained models · COMP6 ANI-MD | energy MAE: 249.7 -> 69.9 | Improved |
| remove interaction layers (ablation) | TensorNet · rMD17 toluene | energy MAE: 0.33 -> 3.3 | Hurt |
| add one interaction layer instead of none (ablation) | TensorNet · rMD17 aspirin | energy MAE: 9.8 -> 2.7 | Improved |
| add a second interaction layer (ablation) | TensorNet 1L, O(3), cutoff 4.5Å · rMD17 toluene | energy MAE: 0.33 -> 0.26 | Improved |
| use SO(3) instead of O(3) equivariance (ablation) | TensorNet 1L, cutoff 4.5Å · rMD17 aspirin | force MAE: 10.1 -> 13.1 | Hurt |
| increase cutoff from 4.5Å to 9Å (ablation) | TensorNet 0L, O(3) · rMD17 toluene | energy MAE: 3.3 -> 2.0 | Improved |
| increase cutoff from 4.5Å to 9Å (ablation) | TensorNet 2L, O(3) · rMD17 toluene | energy MAE: 0.26 -> 0.32 | Hurt |
| predict energy, forces, dipole, polarizability, and shielding jointly | TensorNet · Ethanol in vacuum | dipole MAE: 0.003(0) vs 0.003 | No clear effect |
| use TensorNet instead of FieldSchNet | ethanol property prediction · Ethanol in vacuum | energy MAE: 0.017 -> 0.008(1) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MACE — MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields (NeurIPS 2022) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing (NeurIPS 2024 · similarity 0.92)
- FlashTP: Fused, Sparsity-Aware Tensor Product for Machine Learning Interatomic Potentials (ICML 2025 · similarity 0.89)
- Efficient and Scalable Density Functional Theory Hamiltonian Prediction through Adaptive Sparsity (ICML 2025 · similarity 0.87)
- E2Former: An Efficient and Equivariant Transformer with Linear-Scaling Tensor Products (NeurIPS 2025 · similarity 0.87)
- The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains (NeurIPS 2024 · similarity 0.87)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
