Research map / Graph learning / Graph algorithms and clustering
TpuGraphs: A Performance Prediction Dataset on Large Tensor Computational Graphs
Phitchaya Mangpo Phothilimthana, Sami Abu-El-Haija, Kaidi Cao, Bahare Fatemi, Michael Burrows, Charith Mendis, Bryan Perozzi
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 |
|---|---|---|---|
| train with ListMLE ranking loss instead of MSE (ablation) | EarlySAGE · Tile:XLA test | top-1 error %: 22.0 -> 9.1 | Improved |
| train with ListMLE ranking loss instead of MSE (ablation) | EarlyGCN · Tile:XLA test | top-1 error %: 21.0 -> 11.4 | Improved |
| train with ListMLE ranking loss instead of MSE (ablation) | LateGCN · Tile:XLA test | top-1 error %: 67.0 -> 17.4 | Improved |
| early-join config features with node features instead of late-join (ablation) | SAGE, ListMLE · Tile:XLA test | top-1 error %: 71.8 -> 9.1 | Improved |
| early-join instead of late-join (ablation) | GCN, ListMLE · Tile:XLA test | top-1 error %: 17.4 -> 11.4 | Improved |
| use GraphSAGE instead of GCN (ablation) | EarlyJoin ListMLE · Tile:XLA test | top-1 error %: 11.4 -> 9.1 | No clear effect |
| remove GNN (MLP baseline) (ablation) | ListMLE · Tile:XLA test | top-1 error %: 9.1 -> 52.9 | Hurt |
| train on full graph instead of graph segments (GST) (ablation) | layout model · Layout:XLA:Random test | Kendall tau: 0.32 -> 0.37 | Improved |
| train on full graph instead of graph segments (GST) (ablation) | layout model · Layout:XLA:Random test | top-1 error %: 8.1 -> 12.2 | Hurt |
| MSE loss instead of ranking loss (ablation) | layout model · Layout:XLA:Random test | Kendall tau: 0.32 -> 0.31 | No clear effect |
| default search space instead of random search space | layout model · Layout:XLA test | Kendall tau: 0.34 -> 0.21 | Hurt |
| XLA graphs instead of NLP graphs | layout model · Layout:Random test | Kendall tau: 0.53 -> 0.34 | Hurt |
Closest papers by meaning
- TenSet: A Large-scale Program Performance Dataset for Learned Tensor Compilers (NeurIPS 2021 · similarity 0.87)
- PERFOGRAPH: A Numerical Aware Program Graph Representation for Performance Optimization and Program Analysis (NeurIPS 2023 · similarity 0.87)
- Surprisingly Strong Performance Prediction with Neural Graph Features (ICML 2024 · similarity 0.85)
- Einsum Benchmark: Enabling the Development of Next-Generation Tensor Execution Engines (NeurIPS 2024 · similarity 0.85)
- GraphLand: Evaluating Graph Machine Learning Models on Diverse Industrial Data (NeurIPS 2025 · similarity 0.84)
- tnGPS: Discovering Unknown Tensor Network Structure Search Algorithms via Large Language Models (LLMs) (ICML 2024 · similarity 0.83)
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.
