Research map / Vision-language and multimodal / Robot action and navigation
CURE: Cumulative Knowledge Reuse for Efficient Device-Server Hybrid Inference in Vision-Language Models
Youngjun Lee, Doyoung Kim, Junhyeok Kang, Hwanjun Song, Jae-Gil Lee
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 server-side predictions to the reservoir update (ablation) | CURE · 13 datasets (avg) | top-1 accuracy (%): 78.26 -> 85.75 | Improved |
| add server-side predictions to the reservoir update (ablation) | CURE · 13 datasets (avg) | delegation frequency (%): 37.15 -> 33.26 | Improved |
| add on-device confident predictions to the reservoir update (ablation) | CURE · 13 datasets (avg) | delegation frequency (%): 39.55 -> 33.26 | Improved |
| add on-device confident predictions to the reservoir update (ablation) | CURE · 13 datasets (avg) | top-1 accuracy (%): 85.48 -> 85.75 | Improved |
| add knowledge fusion of VLM and reservoir predictions (ablation) | CURE · 13 datasets (avg) | top-1 accuracy (%): 83.01 -> 85.75 | Improved |
| use shared covariance instead of class-wise covariance in the reservoir (ablation) | CURE · 13 datasets (avg) | top-1 accuracy (%): 85.75 -> 85.71 | No clear effect |
| use shared covariance instead of class-wise covariance in the reservoir (ablation) | CURE · 13 datasets (avg) | delegation frequency (%): 33.26 -> 33.51 | No clear effect |
| add cumulative knowledge reuse (reservoir) over non-cumulative hybrid | Non-Cumulative Hybrid · 13 datasets (avg) | delegation frequency (%): 74.92 -> 33.26 | Improved |
| add cumulative knowledge reuse over always-delegate | Always-Delegate · 13 datasets (avg) | top-1 accuracy (%): 84.70 -> 85.75 | No clear effect |
| add cumulative knowledge reuse over non-cumulative hybrid | Non-Cumulative Hybrid · Stanford Cars | end-to-end runtime (s): 1848.71 -> 659.64 | Improved |
| add cumulative knowledge reuse over always-delegate | Always-Delegate · Stanford Cars | end-to-end runtime (s): 2421.73 -> 659.64 | Improved |
| use device-server delegation instead of on-device-only test-time adaptation | DOTA · 13 datasets (avg) | top-1 accuracy (%): 80.63 -> 85.75 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ADAPT — ADAPT: Vision-Language Navigation With Modality-Aligned Action Prompts (CVPR 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VLsI: Verbalized Layers-to-Interactions from Large to Small Vision Language Models (CVPR 2025 · similarity 0.87)
- Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models (ICML 2026 · similarity 0.87)
- Enabling VLA Action Self-Verification via VLM Token Probability Bucketing (NeurIPS 2026 · similarity 0.86)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (ICLR 2025 · similarity 0.86)
- DOTA: Distributional Test-time Adaptation of Vision-Language Models (NeurIPS 2025 · similarity 0.86)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (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.
