Belief Dynamics Prediction on Custom Dataset Belief Dynamics 1.0 (test)
39.7Macro-Avg Precision (m1)Ours-full
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ours-fullmodel_name=Hierarchical energy-based model, optimizer=Adam2021.04 | 39.7 | 41.5 | 31.6 | 31.5 | 27.8 | 34.4 | 43.1 | 44.3 | 35.1 | 34.9 | 29.9 | 37.5 | |
| Uniform eventablation_variant=uniform event posterior distribution2021.04 | 38.5 | 41.3 | 29.3 | 31 | 26.7 | 33.4 | 36.3 | 36.6 | 30.4 | 32.8 | 27.8 | 32.8 | |
| Event priorablation_variant=only event prior for event assignment2021.04 | 38.4 | 40.9 | 29.4 | 30.7 | 26.4 | 33.2 | 36.5 | 36.4 | 30.5 | 32.4 | 27.3 | 32.6 | |
| Init. seg.ablation_variant=unsupervised interactive segments without temporal re-clustering2021.04 | 37.1 | 41.8 | 29.3 | 30.1 | 26.5 | 33 | 34.6 | 36.6 | 30.2 | 31.4 | 27.4 | 32 | |
| CNN w/ HOG & memoryarchitecture=CNN, features=HOG, memory_context=true2021.04 | 27.7 | 27.9 | 26.6 | 26.7 | 25.9 | 27 | 28.5 | 28.5 | 24.6 | 25 | 15.5 | 24.4 | |
| Features w/ memoryarchitecture=MLP, memory_context=true2021.04 | 27.2 | 27.8 | 25.3 | 26 | 25.6 | 26.4 | 27.4 | 28.8 | 23 | 22.7 | 19.1 | 24.2 | |
| Chancetype=random baseline2021.04 | 25 | 25 | 25 | 25 | 25 | 25 | 10.3 | 10.4 | 10.2 | 10.1 | 10 | 10.2 | |
| CNNarchitecture=CNN baseline2021.04 | 25 | 25 | 25 | 25 | 25 | 25 | 17.1 | 16.7 | 16.9 | 17.4 | 25 | 18.6 | |
| CNN w/ HOG-LSTMarchitecture=CNN, features=HOG, temporal_module=LSTM2021.04 | 25 | 25 | 25 | 25 | 25 | 25 | 16.7 | 13.2 | 20.5 | 18.2 | 25 | 18.7 |