Research map / Architectures and efficiency / Pruning and efficient tuning
XTransfer: Modality-Agnostic Few-Shot Model Transfer for Human Sensing at the Edge
Yu Zhang, Xi Zhang, Hualin zhou, Xinyuan Chen, Shang Gao, Hong Jia, Jianfei Yang, Yuankai Qi, Tao Gu
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 LWS layer-wise search on top of SRR (Our-Single vs SRR) (ablation) | XTransfer · HHAR (5-shot) | accuracy (%): 61.8 -> 71.8 | Improved |
| add layer channel removal to SRR (ablation) | XTransfer SRR · HHAR (5-shot) | accuracy (%): 63.6 -> 61.8 | Mixed |
| add feature space alignment (class pairing) to repair loss (ablation) | XTransfer repair · HHAR initial-layer repair | speed of repair loss decrease: 1.87 times faster | Improved |
| use PCA anchor space with repair loss (ablation) | XTransfer repair loss · HHAR (5-shot) | accuracy (%): 69.33 -> 71.85 | Improved |
| use repair loss with PCA instead of MMD loss (ablation) | XTransfer · HHAR (3/5/10-shot average) | accuracy (%): 65.89 -> 71.85 | Improved |
| use standard PCA instead of sparse PCA (ablation) | XTransfer · HHAR (5-shot) | accuracy (%): 64.67 -> 71.85 | Improved |
| use PCA-#2 instead of PCA-Max (ablation) | XTransfer · HHAR (3/5/10-shot) | accuracy (%): average drop of 4.5% for PCA-Max vs PCA-80% | Improved |
| use Repair loss instead of N-Pair and Triplet loss (ablation) | XTransfer · HHAR (5-shot) | S-score: 0.2 vs 0.05 and 0.07 | Improved |
| use proposed layer channel removal instead of L2-based pruning (ablation) | XTransfer · HHAR (5-shot) | average S-score: 0.1 average S-score | Improved |
| add pre-search check and dynamic search range (Multi-Efficient vs Multi-Pre) (ablation) | XTransfer Multi-ResNet18 search · HHAR | accuracy and search time: reduces search time by 2.1 to 4 times | Improved |
| use only pre-search check (Multi-Pre) instead of Multi (ablation) | XTransfer Multi-ResNet18 search · HHAR | accuracy (%): degraded accuracy by 3.36% and 7.94% | Hurt |
| use Image source instead of Text, Audio or Sensing source models (ablation) | XTransfer · HHAR (3- to 10-shot) | accuracy (%): 70.5% vs 69.6%, 64.3%, 67% | Improved |
Closest papers by meaning
- X-Fi: A Modality-Invariant Foundation Model for Multimodal Human Sensing (ICLR 2025 · similarity 0.87)
- Sample- and Parameter-Efficient Auto-Regressive Image Models (CVPR 2025 · similarity 0.82)
- Implicit Modeling for Transferability Estimation of Vision Foundation Models (NeurIPS 2025 · similarity 0.81)
- ModHiFi: Identifying High Fidelity predictive components for Model Modification (NeurIPS 2025 · similarity 0.81)
- DeepCompress-ViT: Rethinking Model Compression to Enhance Efficiency of Vision Transformers at the Edge (CVPR 2025 · similarity 0.81)
- Stepping Forward on the Last Mile (NeurIPS 2024 · similarity 0.81)
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.
