Research map / Optimization / Distributed and compressed
Distributed Methods with Compressed Communication for Solving Variational Inequalities, with Theoretical Guarantees
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 MASHA1 or MASHA2 instead of QGD or error feedback | Distributed bilinear saddle-point problem · Bilinear saddle-point problem | convergence speed | Helped |
| Compress differences between operator evaluations rather than applying compressed Extra Gradient | MASHA1 and MASHA2 · Bilinear saddle-point problem | convergence speed | Helped |
| Use Top 30% contractive compression instead of Random 30% unbiased compression | MASHA2 versus MASHA1 · Bilinear saddle-point problem | convergence | Helped |
| Use full 8-bit quantization | ALBERT-large adversarial training · BookCorpus and Wikipedia; downstream performance after approximately 80 billion tokens | training loss | Hurt |
| Use mixed 8-bit quantization instead of uncompressed gradients | ALBERT-large adversarial training · BookCorpus and Wikipedia; downstream tasks after approximately 80 billion tokens | downstream performance | No clear effect |
| Use Power compression with error feedback instead of uncompressed gradients | ALBERT-large adversarial training · BookCorpus and Wikipedia; downstream tasks after approximately 80 billion tokens | downstream performance | Hurt |
| Apply gradient compression, especially Power compression | ALBERT-large adversarial training · Communication rounds with 16 workers | communication time | Helped |
Closest papers by meaning
- Similarity, Compression and Local Steps: Three Pillars of Efficient Communications for Distributed Variational Inequalities (NeurIPS 2023 · similarity 0.95)
- Communication-Efficient Gradient Descent-Accent Methods for Distributed Variational Inequalities: Unified Analysis and Local Updates (ICLR 2024 · similarity 0.89)
- Accelerated Methods with Compressed Communications for Distributed Optimization Problems Under Data Similarity (AAAI 2025 · similarity 0.89)
- Scalable Distributed Stochastic Optimization via Bidirectional Compression: Beyond Pessimistic Limits (NeurIPS 2026 · similarity 0.88)
- Towards Faster Decentralized Stochastic Optimization with Communication Compression (ICLR 2025 · similarity 0.87)
- Distributed Online Convex Optimization with Compressed Communication: Optimal Regret and Applications (NeurIPS 2026 · similarity 0.87)
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.