Research map / Label-efficient and robust learning / Knowledge distillation
MAGIC: Meta-Ability Guided Interactive Chain-of-Distillation for Effective-and-Efficient Vision-and-Language Navigation
Liuyi Wang, Zongtao He, Mengjiao Shen, Jingwei Yang, Chengju Liu, Qijun Chen
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 meta-ability knowledge distillation (MAKD) (ablation) | MAGIC-B · R2R val-unseen | SR: 74.88 -> 76.33 | Improved |
| add meta-ability knowledge distillation (MAKD) (ablation) | MAGIC-B · R2R val-unseen | SPL: 64.82 -> 67.86 | Improved |
| add global topological cross-modal location knowledge transfer (ablation) | MAGIC-B MAKD · R2R val-unseen | SR: 73.95 -> 76.33 | Improved |
| add behavior decision-making knowledge transfer (ablation) | MAGIC-B MAKD · R2R val-unseen | OSR: 84.89 -> 83.61 | Hurt |
| add text knowledge transfer (ablation) | MAGIC-B MAKD · R2R val-unseen | SR: 75.78 -> 76.33 | Improved |
| add transfer of hidden features to attention and logits KD (ablation) | MAGIC-B MAKD · R2R val-unseen | SR: SR drops 1.9% | Improved |
| use randomization weighting (MKRW) instead of EqualAdd, LearnedWeights, GradAdjust (ablation) | MAKD losses · R2R val-unseen | SR and SPL | Improved |
| add teacher determination weighting (beta=0.7) instead of beta=0 (ablation) | MKTD · R2R val-unseen | SPL: SPL drops 1.6% at beta=0 | Improved |
| increase MKTD attenuation beta to 0.9 (ablation) | MKTD · R2R val-unseen | SPL: SPL reduces 4.5% | Hurt |
| add interactive co-training (ICoD) with student feedback (ablation) | MAGIC-S · R2R val-unseen | SR: 73.44 -> 76.03 | Improved |
| add interactive co-training (ICoD) (ablation) | MAGIC-L teacher · R2R val-unseen | SR: 77.82 -> 78.76 | Improved |
| continue training teacher without student feedback (ablation) | teacher model · R2R | generalization | Hurt |
Closest papers by meaning
- NavCoT: Boosting LLM-Based Vision-and-Language Navigation via Learning Disentangled Reasoning (TPAMI 2025 · similarity 0.88)
- EvolveNav: Empowering LLM-Based Vision-Language Navigation via Self-Improving Embodied Reasoning (TPAMI 2026 · similarity 0.87)
- Vision-Language Navigation with Energy-Based Policy (NeurIPS 2024 · similarity 0.87)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.87)
- Human-Aware Vision-and-Language Navigation: Bridging Simulation to Reality with Dynamic Human Interactions (NeurIPS 2024 · similarity 0.86)
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · 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.
