Research map / Label-efficient and robust learning / Recognition and long-tail
Scaling-Aware Data Selection for End-to-End Autonomous Driving Systems
Tolga Dimlioglu, Nadine Chang, Maying Shen, Rafid Mahmood, Jose M. Alvarez
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 |
|---|---|---|---|
| use scaling-aware mixture selection (MOSAIC) instead of random selection | E2E planner · OpenScene, 250 clips | EPDMS: 72.84 -> 77.38 | Improved |
| use MOSAIC instead of Coreset selection | E2E planner · Navtrain, 1600 clips | EPDMS: 89.30 -> 90.18 | Improved |
| use uncertainty-based selection instead of random selection | E2E planner · OpenScene, 4000 clips | EPDMS: 80.38 -> 73.46 | Hurt |
| cluster the pool by captions instead of geolocation for Chameleon | Chameleon · Navtrain, 1600 clips | EPDMS: 89.50 -> 86.99 | Hurt |
| cluster the pool by captions instead of geolocation for MOSAIC | MOSAIC · Navtrain, 1600 clips | EPDMS: 90.18 -> 89.98 | No clear effect |
| add clustering with scaling-aware allocation (versus w/o Clustering) (ablation) | MOSAIC · OpenScene | EPDMS | Improved |
| add importance ranking within clusters (versus w/o Ranking) (ablation) | MOSAIC · OpenScene, up to 800 clips | EPDMS | Improved |
| replace EPDMS-based ranking with cheap signals (loss, gradient norm, sensitivity) (ablation) | MOSAIC ranking · OpenScene | Kendall-Tau correlation with EPDMS ranking | Hurt |
| use the linear separable sum of cluster scaling fits to predict performance | MOSAIC · Navtrain, 1600 clips | EPDMS: Actual 90.2, Estimated 91.1 | No clear effect |
Closest papers by meaning
- TMCOSS: Thresholded Multi-Criteria Online Subset Selection for Data-Efficient Autonomous Driving (ICCV 2021 · similarity 0.86)
- Den-TP: A Density-Balanced Data Curation and Evaluation Framework for Trajectory Prediction (CVPR 2026 · similarity 0.85)
- SubjectDrive: Scaling Generative Data in Autonomous Driving via Subject Control (AAAI 2025 · similarity 0.84)
- ActiveAD: Planning-Oriented Active Learning for End-to-End Autonomous Driving (CVPR 2026 · similarity 0.84)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.84)
- Continuously Learning, Adapting, and Improving: A Dual-Process Approach to Autonomous Driving (NeurIPS 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.
