Research map / Vision-language and multimodal / Robot action and navigation
FT-NCFM: An Influence-Aware Data Distillation Framework for Efficient VLA Models
Kewei Chen, Yayu Long, Shuai Li, Mingsheng Shang
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 contrastive verification refinement to influence weights (ablation) | FT-NCFM (5% data) · CALVIN ABC->D | Avg. Len: 2.81 -> 3.11 | Improved |
| add FT influence engine instead of random weights (ablation) | FT-NCFM (5% data) · CALVIN ABC->D | Avg. Len: 2.15 -> 3.11 | Improved |
| synthesize coreset with FT-NCFM instead of random sampling at 5% | VLA training data · CALVIN ABC->D | Avg. Len: 1.88 -> 3.11 | Improved |
| synthesize coreset with FT-NCFM instead of influence-function selection at 5% | VLA training data · CALVIN ABC->D | Avg. Len: 2.45 -> 3.11 | Improved |
| increase synthetic data ratio from 1% to 10% (ablation) | FT-NCFM · LIBERO (avg of 4 suites) | success rate (%): 46.9 -> 74.2 | Improved |
| data distillation (10%) instead of policy distillation on full data | VLA training · CALVIN ABC->D | Avg. Len / training time: 3.20 -> 3.47 Avg. Len; 178 -> 31.5 GPU-h | Improved |
| train on 10% distilled data versus full data | FT-NCFM vs SpatialVLA · LIBERO (avg of 4 suites) | success rate (%): 78.1 -> 74.2 | Hurt |
Closest papers by meaning
- Enabling VLA Action Self-Verification via VLM Token Probability Bucketing (NeurIPS 2026 · similarity 0.87)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 2026 · similarity 0.86)
- Efficient Data Selection at Scale via Influence Distillation (NeurIPS 2025 · similarity 0.85)
- What Can RL Bring to VLA Generalization? An Empirical Study (NeurIPS 2025 · similarity 0.85)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.85)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.85)
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.
