Research map / Probabilistic and scientific modeling / Molecules and proteins
DECODE: Domain-Aware Continual Domain Expansion for Motion Prediction
Boqi Li, Haojie Zhu, Henry X. Liu
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 deep Bayesian uncertainty balancing of generalized and specialized models (ablation) | DECODE after phase 1 · HighD minADE | minADE: 65.87 -> 0.839 | Improved |
| add deep Bayesian uncertainty balancing (ablation) | DECODE after phase 1 · InD minADE | minADE: 1.984 -> 1.165 | Improved |
| add deep Bayesian uncertainty balancing (ablation) | DECODE after phase 1 · RounD minADE | minADE: 0.527 -> 0.531 | No clear effect |
| increase prior evidence e0 beyond its best value (ablation) | DECODE · RounD | minADE/minFDE | Hurt |
| increase prior evidence e0 (ablation) | DECODE · HighD and InD | minADE/minFDE | Improved |
| use DECODE hypernetwork instead of rehearsal DER++(500) | DECODE · motion prediction continual learning | AER minADE: 0.598 -> 0.584 | Improved |
| use DECODE instead of DualPrompt | DECODE · motion prediction continual learning | FGT minADE: 0.040 -> 0.044 | Mixed |
| replace regularization (EWC) with rehearsal (ER) | baselines · motion prediction continual learning | AER minADE: 2.038 -> 0.896 | Improved |
| place DualPrompt E-Prompts in encoder and decoder instead of encoder only (ablation) | DualPrompt · motion prediction | AER minADE: 0.794 -> 0.608 | Improved |
| place CODA-P prompts in decoder only instead of encoder+decoder (ablation) | CODA-P · motion prediction | AER minADE: 0.685 -> 0.647 | Improved |
| apply DECODE to ViT-B/16 on 5-datasets vs DualPrompt | DECODE · 5-datasets | average accuracy: 88.08 -> 90.56 | Improved |
| apply DECODE to ViT-B/16 on CIFAR-100 vs CODA-P | DECODE · 10-task CIFAR-100 | average accuracy: 91.71 -> 86.41 | Hurt |
Closest papers by meaning
- Continual Learning for Motion Prediction Model via Meta-Representation Learning and Optimal Memory Buffer Retention Strategy (CVPR 2024 · similarity 0.88)
- Generalized Predictive Model for Autonomous Driving (CVPR 2024 · similarity 0.87)
- Motion Forecasting in Continuous Driving (NeurIPS 2024 · similarity 0.86)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.86)
- Uncertainty-Guided Never-Ending Learning to Drive (CVPR 2024 · similarity 0.86)
- UniMotion: A Unified Motion Framework for Simulation, Prediction and Planning (NeurIPS 2025 · similarity 0.86)
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.
