pT estimation on CMS Trigger Dataset
0.0005LossGNN (each station as a node) - 4 MPL (config 1)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GNN (each station as a node) - 4 MPL (config 1)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=MSE2025.07 | 0.0005 | 52.6406 | 101,152 | |
| GNN (each station as a node) - GCN, 2 GAT, SAGEEdges=Sequential (0 → 1 → 2 → 3), Model Backbone=GCN, 2 GAT, SAGE (embed dim: 128), Loss Function=pT2025.07 | 0.0007 | 25.4295 | 59,812 | |
| GNN (each station as a node) - 4 MPL (config 3)Edges=Sequential (0 → 1 → 2 → 3), Model Backbone=4 MPL (embed dim: 128), Loss Function=MSE2025.07 | 0.0019 | 13.3034 | 101,152 | |
| GNN (each station as a node) - 4 MPL (config 2)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=MSE2025.07 | 0.0026 | 13.5296 | 101,152 | |
| GNN (each station as a node) - 4 MPL (config 4)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=pT2025.07 | 0.0042 | 2.0716 | 101,152 | |
| GNN (each station as a node) - 2 EdgeConvEdges=Fully-Connected, Model Backbone=2 EdgeConv (embed dim: 64), Loss Function=MSE2025.07 | 0.1181 | 0.8217 | 19,649 | |
| GNN (each station as a node) - 4 EdgeConv (Custom pT)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=Custom pT2025.07 | 0.1273 | 0.8525 | 3,005 | |
| GNN (η value as a node) - 4 MPL (embed dim: 28)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 28), Loss Function=MSE2025.07 | 0.1494 | 0.9416 | 6,545 | |
| GNN (each station as a node) - 4 EdgeConv (MSE)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=MSE2025.07 | 0.1518 | 0.9481 | 3,005 | |
| GNN (η value as a node) - 4 MPL (MSE, 5903 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.1631 | 0.9921 | 5,903 | |
| GNN (η value as a node) - 2 MPLEdges=Fully-Connected, Model Backbone=2 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.2068 | 1.1333 | 6,112 | |
| GNN (η value as a node) - 4 MPL (MSE, 5579 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.209 | 1.1469 | 5,579 | |
| GNN (η value as a node) - 4 MPL (MSE, 6437 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.2113 | 1.1457 | 6,437 | |
| GNN (bending angle as a node) - 2 MPLEdges=Fully-Connected, Model Backbone=2 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.2416 | 1.2359 | 6,112 | |
| GNN (bending angle as a node) - 4 MPL (MSE, 6437 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.2432 | 1.2421 | 6,437 | |
| GNN (bending angle as a node) - 4 MPL (embed dim: 28)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 28), Loss Function=MSE2025.07 | 0.2475 | 1.2509 | 6,545 | |
| GNN (bending angle as a node) - 4 MPL (MSE, 5903 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 0.2519 | 1.2642 | 5,903 | |
| GNN (each station as a node) - 4 MPL (config 6)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=Custom pT2025.07 | 0.4529 | 0.8129 | 101,152 | |
| GNN (each station as a node) - 4 MPL (config 5)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=Custom pT2025.07 | 0.468 | 0.8178 | 101,152 | |
| GNN (each station as a node) - 4 MPL (config 7)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=Custom pT2025.07 | 0.4876 | 0.8923 | 101,152 | |
| GNN (bending angle as a node) - 4 EdgeConv (Custom pT)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=Custom pT2025.07 | 0.4982 | 1.7798 | 2,909 | |
| GNN (bending angle as a node) - 4 EdgeConv (MSE)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=MSE2025.07 | 0.4997 | 1.7859 | 2,909 | |
| GNN (η value as a node) - 4 EdgeConv (MSE)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=MSE2025.07 | 0.994 | 2.9878 | 2,909 | |
| GNN (η value as a node) - 4 EdgeConv (Custom pT)Edges=Fully-Connected, Model Backbone=4 EdgeConv (embed dim: 16), Loss Function=Custom pT2025.07 | 0.994 | 2.9896 | 2,909 | |
| GNN (each station as a node) - 4 MPLEdges=Fully-Connected, Model Backbone=4 MPL (embed dim: 64), Loss Function=pT2025.07 | 1.3384 | 16.2277 | 101,487 | |
| GNN (each station as a node) - 4 MPLEdges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=pT2025.07 | 1.3649 | 15.8229 | 101,152 | |
| TabNetModeling Approach=TabNet, Loss Function=MSE2025.07 | 2.9746 | 0.9607 | 6,696 | |
| GNN (each feature as a node) - 4 MPL (101k)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=pT2025.07 | 3.7945 | 41.7086 | 99,952 | |
| GNN (each station as a node) - 4 GCNEdges=Sequential (0 → 1 → 2 → 3), Model Backbone=4 GCN (embed dim: 128), Loss Function=pT2025.07 | 3.8392 | 43.2106 | 55,460 | |
| GNN (bending angle as a node) - 4 MPL (MSE, 5579 params)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 24), Loss Function=MSE2025.07 | 4.0059 | 1.2741 | 5,579 | |
| GNN (each feature as a node) - 4 MPL (3k)Edges=Fully-Connected, Model Backbone=4 MPL (embed dim: 128), Loss Function=pT2025.07 | 4.0914 | 44.7491 | 3,005 | |
| GNN (each feature as a node) - 4 GCNEdges=0 → 1 → 2 → 3 || 2 ↔ {0, 4, 5, 6}, 6, Model Backbone=4 GCN (embed dim: 128), Loss Function=pT2025.07 | 4.3879 | 45.3396 | 55,076 | |
| LSTMModel Backbone=2 LSTM, 1 FCN, Loss Function=MSE2025.07 | 17,629.877 | 16.6884 | 32,257 | |
| FCNNModel Backbone=5 Hidden FCN, Loss Function=MSE2025.07 | 18,647.6816 | 19.3611 | 213,761 | |
| Dual Channel CNN-FCNNModel Backbone=3 Conv2D (Channel-1), 4 FCN (Channel-2), Loss Function=MSE2025.07 | 18,771.123 | 21.9292 | 52,193 | |
| CNN-GridModel Backbone=3 Conv2D, 3 FCN, Loss Function=MSE2025.07 | 18,777.416 | 20.6122 | 230,593 | |
| CNNModel Backbone=6 Conv2D, Loss Function=MSE2025.07 | 18,788.1523 | 18.7456 | 1,050,305 |