Research map / Architectures and efficiency / RNNs and normalization
Short-Dot: Computing Large Linear Transforms Distributedly Using Coded Short Dot Products
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 |
|---|---|---|---|
| use Short-Dot (K=18) instead of MDS coding (K=10) | distributed linear transform · MNIST 10000 dot products, P=20 | mean computation time: 15.3411 -> 10.4306 | Helped |
| use Short-Dot (K=18) instead of uncoded (K=20) | distributed linear transform · MNIST 10000 dot products, P=20 | mean computation time: 11.8653 -> 10.4306 | Helped |
| use uncoded instead of MDS | distributed linear transform · MNIST 10000 dot products, P=20 | mean computation time: 15.3411 -> 11.8653 | Helped |
| Short-Dot versus uncoded variability | distributed linear transform · MNIST, P=20 | maximum computation time: 27.0818 -> 11.8340 | Helped |
| Short-Dot versus uncoded variability | distributed linear transform · MNIST, P=20 | STDEV of computation time: 2.8427 -> 0.9253 | Helped |
| Short-Dot versus MDS spread | distributed linear transform · MNIST, P=20 | STDEV of computation time: 0.8987 -> 0.9253 | No clear effect |
| choose M=P/log P regime with Short-Dot instead of MDS, repetition, uncoded | expected time analysis · shifted-exponential model | expected computation time | Helped |
| Short-Dot versus uncoded/repetition when M linear in P | expected time analysis · shifted-exponential model | expected computation time | Helped |
| Short-Dot versus repetition/MDS when M sub-linear in P | expected time analysis · shifted-exponential model | expected computation time | Helped |
| Short-MDS versus Short-Dot when s does not divide N | straggler resilience K · worst-case K | processors to wait for | Hurt |
Closest papers by meaning
- Coded Computing for Resilient Distributed Computing: A Learning-Theoretic Framework (NeurIPS 2024 · similarity 0.86)
- Leveraging partial stragglers within gradient coding (NeurIPS 2024 · similarity 0.86)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.85)
- Linear Transformers Implicitly Discover Unified Numerical Algorithms (NeurIPS 2025 · similarity 0.84)
- CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs (NeurIPS 2026 · similarity 0.84)
- Approximate Gradient Coding for Distributed Learning with Heterogeneous Stragglers (NeurIPS 2025 · similarity 0.84)
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.