Research map / Federated learning and privacy / Communication-efficient FL
No Retraining at Edge: Efficient Resource-Aware Mixed-Precision Quantization via Federated Supernet Learning
Lianbo Ma, Yonghui Su, Nan Li, Xingwei Wang
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 |
|---|---|---|---|
| apply layer-wise semantic aggregation weights instead of hybrid (ablation) | DFMPQ SAHA · Tiny-ImageNet (alpha=0.5) | test accuracy (%): 66.65 -> 63.70 | Hurt |
| use hybrid semantic-aware aggregation instead of FedAvg (ablation) | DFMPQ · CIFAR-100 (alpha=0.5) | test accuracy (%): 70.69 -> 70.88 | Improved |
| use hybrid semantic-aware aggregation instead of FedAvg (ablation) | DFMPQ · CIFAR-10 (alpha=0.5) | test accuracy (%): 90.42 -> 90.71 | No clear effect |
| use hybrid instead of stage-wise aggregation (ablation) | DFMPQ · CIFAR-10 (alpha=0.5) | test accuracy (%): 90.92 -> 90.71 | No clear effect |
| add CCSA regularization (lambda=0 -> 0.1) (ablation) | DFMPQ mixed-precision ResNet18 · CIFAR-10, CIFAR-100, Tiny-ImageNet (alpha=0.5) | test accuracy | Improved |
| increase CCSA lambda above 0.1 (ablation) | DFMPQ · CIFAR-10, CIFAR-100, Tiny-ImageNet (alpha=0.5) | test accuracy | Hurt |
| use 3-bit mixed-precision supernet instead of retrained quantized baselines | DFMPQ vs FedMPQ · CIFAR-10 (alpha=0.5) | test accuracy (%): 67.10 -> 80.24 | Improved |
| 3-bit mixed-precision DFMPQ versus full-precision FedWS | DFMPQ · CIFAR-10 (alpha=0.5) | test accuracy (%): 91.0 -> 80.24 | Hurt |
| increase number of clients from 10 to 40 | DFMPQ · CIFAR-100 (alpha=0.5) | test accuracy (%): 64.15 -> 36.64 | Hurt |
Closest papers by meaning
- Mixed-Precision Quantization for Federated Learning on Resource-Constrained Heterogeneous Devices (CVPR 2024 · similarity 0.92)
- LBI-FL: Low-Bit Integerized Federated Learning with Temporally Dynamic Bit-Width Allocation (ICML 2025 · similarity 0.87)
- Fed-QSSL: A Framework for Personalized Federated Learning under Bitwidth and Data Heterogeneity (AAAI 2024 · similarity 0.86)
- SDP4Bit: Toward 4-bit Communication Quantization in Sharded Data Parallelism for LLM Training (NeurIPS 2024 · similarity 0.85)
- Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.85)
- JointSQ: Joint Sparsification-Quantization for Distributed Learning (CVPR 2024 · 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.
