Research map / Optimization / Distributed and compressed
Qsparse-local-SGD: Distributed SGD with Quantization, Sparsification and Local Computations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Combine quantization, Top_k sparsification, and error compensation | SGD on ResNet-50 · ImageNet | communication bits to achieve the same loss or top-1 accuracy: around 1/16 compared with Top_k-SGD; over 1000× less than vanilla SGD | Helped |
| Use QTop_k with a 4 bit quantizer | SGD · ImageNet | communication bits for target loss or test accuracy | Helped |
| Use SignTop_k with a 1 bit sign quantizer | SGD · ImageNet | communication bits for target loss or test accuracy | Helped |
| Enable error compensation for quantization and sparsification | SGD · ImageNet | convergence rate | No clear effect |
| Add local iterations to Qsparse operators (ablation) | Qsparse-local-SGD · ImageNet | communication bits for achieving target loss: factor of 6 to 8 times on average | Helped |
| Add local iterations (ablation) | vanilla SGD and compressed SGD methods · ImageNet | convergence rate | No clear effect |
| Replace the unscaled QTop_k operator with the scaled QTop_k operator (ablation) | Qsparse-local-SGD · ImageNet | performance | No clear effect |
Closest papers by meaning
- QSGD: Communication-Efficient SGD via Gradient Quantization and Encoding (NeurIPS 2017 · similarity 0.92)
- Error Compensated Quantized SGD and its Applications to Large-scale Distributed Optimization (ICML 2018 · similarity 0.90)
- A Linear Speedup Analysis of Distributed Deep Learning with Sparse and Quantized Communication (NeurIPS 2018 · similarity 0.90)
- JointSQ: Joint Sparsification-Quantization for Distributed Learning (CVPR 2024 · similarity 0.88)
- LOSCAR-SGD: Local SGD with Communication-Computation Overlap and Delay-Corrected Sparse Model Averaging (NeurIPS 2026 · similarity 0.88)
- A Quadratic Synchronization Rule for Distributed Deep Learning (ICLR 2024 · similarity 0.88)
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.