Research map / Optimization / Distributed and compressed
Communication-Efficient Distributed Learning via Lazily Aggregated Quantized Gradients
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 |
|---|---|---|---|
| Quantize gradients and skip uploads when gradient innovations are small | Distributed logistic regression on MNIST · MNIST logistic regression | Bit #: 7.08×10^9 -> 1.95×10^7 | Helped |
| Quantize gradients and skip uploads when gradient innovations are small | Distributed logistic regression on MNIST · MNIST logistic regression | Accuracy: 0.9082 -> 0.9082 | No clear effect |
| Quantize gradients and skip uploads when gradient innovations are small | Distributed neural network on MNIST · MNIST neural network | Bit #: 4.07×10^11 -> 4.05×10^10 | Helped |
| Quantize gradients and skip uploads when gradient innovations are small | Distributed neural network on MNIST · MNIST neural network | Accuracy: 0.9433 -> 0.9433 | No clear effect |
| Use LAQ instead of LAG | Distributed logistic regression on MNIST · MNIST logistic regression | Bit #: 5.98×10^8 -> 1.95×10^7 | Helped |
| Use LAQ instead of LAG | Distributed logistic regression on MNIST · MNIST logistic regression | Communication #: 2382 -> 620 | Hurt |
| Use LAQ instead of LAG | Distributed neural network on MNIST · MNIST neural network | Bit #: 1.52×10^11 -> 4.05×10^10 | Helped |
| Use LAQ instead of LAG | Distributed neural network on MNIST · MNIST neural network | Accuracy: 0.9433 -> 0.9433 | No clear effect |
| Use SLAQ instead of QSGD | Mini-batch logistic regression on MNIST · MNIST logistic regression (mini-batch) | Bit #: 7.51×10^8 -> 1.94×10^8 | Helped |
| Use SLAQ instead of QSGD | Mini-batch logistic regression on MNIST · MNIST logistic regression (mini-batch) | Accuracy: 0.9021 -> 0.9018 | Hurt |
| Use SLAQ instead of QSGD | Mini-batch neural network on MNIST · MNIST neural network (mini-batch) | Bit #: 2.03×10^10 -> 1.42×10^10 | Helped |
| Use SLAQ instead of QSGD | Mini-batch neural network on MNIST · MNIST neural network (mini-batch) | Accuracy: 0.9100 -> 0.9107 | Helped |
Closest papers by meaning
- LAG: Lazily Aggregated Gradient for Communication-Efficient Distributed Learning (NeurIPS 2018 · similarity 0.95)
- Double Quantization for Communication-Efficient Distributed Optimization (NeurIPS 2019 · similarity 0.90)
- Communication-efficient Distributed Learning for Large Batch Optimization (ICML 2022 · similarity 0.90)
- A Linear Speedup Analysis of Distributed Deep Learning with Sparse and Quantized Communication (NeurIPS 2018 · similarity 0.90)
- Accelerated Dual Method for Distributed Optimization: An Inexact-Gradient View of Local Updates (ICML 2026 · similarity 0.89)
- Approximate Gradient Coding for Distributed Learning with Heterogeneous Stragglers (NeurIPS 2025 · similarity 0.89)
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.