Research map / Architectures and efficiency / Pruning and efficient tuning
Magnitude-Modulated Equivariant Adapter for Parameter-Efficient Fine-Tuning of Equivariant Graph Neural Networks
Dian Jin, Yancheng Yuan, Xiaoming Tao
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 MMEA scalar gating instead of ELoRA | MACE-OFF · rMD17 Aspirin (50 train samples) | force MAE: 18.3 -> 16.4 | Improved |
| add scalar-channel modulation (ablation) | MMEA · rMD17 Aspirin (50 train samples) | force MAE: 30.5 -> 16.4 | Improved |
| add high-order modulation (ablation) | MMEA · rMD17 Aspirin (50 train samples) | energy MAE: 8.3 -> 7.3 | Improved |
| add high-order modulation (ablation) | MMEA · rMD17 Aspirin (50 train samples) | force MAE: 16.6 -> 16.4 | No clear effect |
| add nonlinear activation (ablation) | MMEA · rMD17 Aspirin (50 train samples) | force MAE: 16.4 -> 16.4 | No clear effect |
| share input-head MLP (reuse) between orders (ablation) | MMEA · rMD17 Aspirin (50 train samples) | energy MAE: 9.2 -> 7.3 | Improved |
| use per-channel (not shared) high-order modulation (ablation) | MMEA · rMD17 Aspirin (50 train samples) | force MAE: 16.6 -> 16.4 | No clear effect |
| use equivariant MMEA instead of conventional Adapter (ablation) | MACE-OFF · rMD17 Aspirin (50 train samples) | force MAE: 26.3 -> 16.4 | Improved |
| use MMEA instead of ELoRA on extreme out-of-distribution target | MACE-MP · rMD17 Aspirin (out-of-distribution, MACE-MP) | force MAE: 31.1 -> 31.9 | Hurt |
| increase MMEA rank from 16 to 32 | MMEA · AcAc 600 K test | energy RMSE: 3.6 -> 3.2 | Improved |
| use MMEA instead of ELoRA at 1200 K | MACE-OFF · 3BPA 1200 K test | force RMSE: 42.0 -> 39.7 | Improved |
Closest papers by meaning
- E2Former: An Efficient and Equivariant Transformer with Linear-Scaling Tensor Products (NeurIPS 2025 · similarity 0.89)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.89)
- On the Generalization of Equivariant Graph Neural Networks (ICML 2024 · similarity 0.89)
- Universally Invariant Learning in Equivariant GNNs (NeurIPS 2025 · similarity 0.89)
- Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks? (NeurIPS 2024 · similarity 0.88)
- Equivariant Spherical Transformer for Efficient Molecular Modeling (NeurIPS 2026 · similarity 0.88)
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.
