Research map / Architectures and efficiency / Quantization
SemanticNN: Compressive and Error-Resilient Semantic Offloading for Extremely Weak Devices
Jiaming Huang, Yi Gao, Fuchang Pan, Renjie Li, Wei Dong
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 SemanticNN instead of offloading raw features | split ResNet-50 · ImageNet-200 with transmission errors | top-1 accuracy (%): 32.84 -> 62.93 | Improved |
| use SemanticNN instead of offloading raw images | split ResNet-50 · CIFAR-100 with transmission errors | top-1 accuracy (%): 45.43 -> 53.96 | Improved |
| use SemanticNN instead of offloading raw features | YOLOv3 · VHR-10 with transmission errors | mAP: 0.68 -> 0.81 | Improved |
| use SemanticNN instead of offloading raw features | YOLOv3 · VOC2007 with transmission errors | mAP: 0.81 -> 0.80 | No clear effect |
| use SemanticNN instead of DeepCOD* and AgileNN* compressed offloading | split ResNet-50 / MobileNetV2 · image classification with transmission errors | top-1 accuracy: +7.11% to 30.88% | Improved |
| use CNN-based SemanticNN instead of Transformer-based DeepSC | split ResNet-50 · ImageNet-200 | transmitted data (KB): 98 -> 24.5 | Improved |
| use CNN-based SemanticNN instead of Transformer-based DeepSC | split ResNet-50 · ImageNet-200 | parameters (KB): 9165 -> 1875 | Improved |
| add two-stage Feature-augmentation Learning (denoising autoencoder pretraining) before task training (ablation) | SemanticNN · image classification | accuracy | Improved |
| add attention module in decoder for dynamic BER (ablation) | SemanticNN · varying BERs | accuracy | Improved |
| set distribution loss weight alpha to 2 (ablation) | SemanticNN · quantization center balance | entropy of center frequency | Improved |
| set XAI loss weight gamma to 1 (ablation) | SemanticNN · XAI loss | loss | Improved |
| use symmetric encoder-decoder instead of asymmetric (E1D3, E1D5) (ablation) | ADJSCC · ResNet-50 offloading | accuracy | Improved |
Closest papers by meaning
- CIAR: Interval-based Collaborative Decoding for Image Generation Acceleration (ICLR 2026 · similarity 0.85)
- When MLLMs Meets Compression Distortion: A Coding Paradigm Tailored to MLLMs (ICLR 2026 · similarity 0.84)
- TinyTrain: Resource-Aware Task-Adaptive Sparse Training of DNNs at the Data-Scarce Edge (ICML 2024 · similarity 0.84)
- Cloud-Device Collaborative Learning for Multimodal Large Language Models (CVPR 2024 · similarity 0.82)
- Rosetta Stone For Unified MLLMs: A Unified Tokenizer to Decipher Understanding and Generation (CVPR 2026 · similarity 0.82)
- Semantic Optimal Transport for Sparse Autoencoder Feature Matching and Circuit Compression (NeurIPS 2026 · similarity 0.82)
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.
