Research map / Graph learning / Graph algorithms and clustering
PACE: A Parallelizable Computation Encoder for Directed Acyclic Graphs
Zehao Dong, Muhan Zhang, Fuhai Li, Yixin Chen
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 PACE instead of DAGNN | DAG encoding and performance prediction · NA | RMSE: 0.264 ± 0.004 -> 0.254 ± 0.002 | Improved |
| Use PACE instead of DAGNN | DAG encoding and performance prediction · BN | RMSE: 0.122 ± 0.004 -> 0.115 ± 0.004 | Improved |
| Use PACE instead of DAGNN | DAG encoding for code token prediction · OGBG-CODE2 | test F1: 0.1751 ± 0.0049 -> 0.1779 ± 0.0021 | Improved |
| Add masked attention to dag2seq + Transformer (ablation) | PACE · NA | RMSE: 0.255 -> 0.254 | Improved |
| Add masked attention to dag2seq + Transformer (ablation) | PACE · BN | RMSE: 0.119 -> 0.115 | Improved |
| Add masked attention to dag2seq + Transformer (ablation) | PACE with NAS101 DNGO search · NAS101 | regret (%): 0.479 -> 0.391 | Improved |
| Add masked attention to dag2seq + Transformer (ablation) | PACE with NAS301 DNGO search · NAS301 | accuracy (%): 94.501 -> 94.507 | Improved |
| Add dag2seq positional encoding and attention mask (ablation) | PACE · NA | RMSE: 0.981 -> 0.254 | Improved |
| Add dag2seq positional encoding and attention mask (ablation) | PACE · BN | RMSE: 0.368 -> 0.115 | Improved |
| Add dag2seq positional encoding and attention mask (ablation) | PACE with NAS101 DNGO search · NAS101 | regret (%): 0.600 -> 0.391 | Improved |
| Add dag2seq positional encoding and attention mask (ablation) | PACE with NAS301 DNGO search · NAS301 | accuracy (%): 94.467 -> 94.507 | Improved |
| Use PACE rather than DAGNN | DAG encoding on OGBG-CODE2 · OGBG-CODE2 | training time (min): 166.19 ± 3.52 -> 40.71 ± 0.19 | Improved |
Later papers that compared with this method
1 later accepted paper reports PACE as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- CktGNN: Circuit Graph Neural Network for Electronic Design Automation (ICLR 2023 · better on 3 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
- D-VAE: A Variational Autoencoder for Directed Acyclic Graphs (NeurIPS 2019 · similarity 0.89)
- Directed Acyclic Graph Neural Networks (ICLR 2021 · similarity 0.89)
- A Structure-Aware Framework for Learning Device Placements on Computation Graphs (NeurIPS 2024 · similarity 0.88)
- Directed Graph Grammars for Sequence-based Learning (ICML 2025 · similarity 0.87)
- LayerDAG: A Layerwise Autoregressive Diffusion Model for Directed Acyclic Graph Generation (ICLR 2025 · similarity 0.86)
- Deep Gaussian Processes on Directed Acyclic Graphs (NeurIPS 2026 · similarity 0.85)
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.
