Research map / Architectures and efficiency / Attention and transformers
Particle Transformer for Jet Tagging
Huilin Qu, Congqiao Li, Sitian Qian
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 |
|---|---|---|---|
| add pairwise interaction term to attention (P-MHA) vs plain MHA (ablation) | ParT · JetClass | Accuracy: +1.2% accuracy (0.861 vs 0.849) | Improved |
| add pairwise interaction term to attention (P-MHA) vs plain MHA (ablation) | ParT · JetClass | Rej50% for t->bqq' and H->4q: background rejection reduced by 20-30% for most signals without it | Improved |
| replace ParticleNet with ParT (P-MHA transformer) | JetClass baseline comparison · JetClass | Accuracy: +1.7% (0.861 vs 0.844) | Improved |
| pre-train ParT on JetClass then fine-tune (ablation) | ParT · Top quark tagging | Rej30%: 2766 vs 1602 from scratch; 70% over ParticleNet | Improved |
| pre-train ParT on JetClass then fine-tune (ablation) | ParT · Quark-gluon tagging (full) | Rej50%: 50.6 vs 47.9 from scratch | Improved |
| pre-train ParticleNet on JetClass then fine-tune (ablation) | ParticleNet · Top quark tagging | Rej30%: only a small improvement over training from scratch | No clear effect |
| increase training set from 2M to 100M jets (ablation) | ParT · JetClass | Accuracy: 0.836 (2M) to 0.861 (100M), 2.5% gain | Improved |
| increase training set from 2M to 100M jets (ablation) | ParticleNet · JetClass | Accuracy: 0.828 (2M) to 0.844 (100M), 1.6% gain | Improved |
Closest papers by meaning
- Patch Hierarchical Attention Transformers for Efficient Particle Jet Tagging (NeurIPS 2026 · similarity 0.90)
- PDE-Transformer: Efficient and Versatile Transformers for Physics Simulations (ICML 2025 · similarity 0.80)
- Jet-Nemotron: Efficient Language Model with Post Neural Architecture Search (NeurIPS 2025 · similarity 0.80)
- Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems (ICML 2025 · similarity 0.79)
- Locality-Sensitive Hashing-Based Efficient Point Transformer with Applications in High-Energy Physics (ICML 2024 · similarity 0.79)
- Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics (NeurIPS 2024 · similarity 0.79)
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.
