Research map / Reinforcement learning / Visual and representation RL
Learning Generalizable Device Placement Algorithms for Distributed Machine Learning
ravichandra addanki, Shaileshh Bojja Venkatakrishnan, Shreyan Gupta, Hongzi Mao, Mohammad Alizadeh
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 |
|---|---|---|---|
| use Placeto iterative-improvement graph-embedding policy instead of RNN-based placer | device placement · NMT 4 GPUs | placement runtime (s): 3.15 -> 2.63 | Improved |
| use Placeto instead of RNN-based placer | device placement · Inception-V3 4 GPUs | placements sampled: 35.8 K -> 5.8 K | Improved |
| use Placeto instead of RNN-based placer | device placement · Inception-V3 2 GPUs | placement runtime (s): 1.17 -> 1.18 | No clear effect |
| use Placeto instead of RNN-based placer | device placement · NMT 4 GPUs | placements sampled: 51.7 K -> 94 K | Hurt |
| use Placeto instead of RNN-based placer | device placement · NASNet 4 GPUs | placement runtime (s): 0.76 -> 0.74 | Improved |
| apply policy zero-shot to unseen graphs with Placeto vs RNN-based | generalization · nmt, ptb, cifar10 test graphs | placement runtime | Improved |
| train on fixed node traversal orders then test on unseen random orders, Placeto vs RNN-based | generalization · Inception-V3 | placement runtime: within 5% vs about 30% worse | Improved |
| replace graph-neural-network policy with simple aggregator (ablation) | Placeto · nmt dataset | placement runtime: about 20% worse | Hurt |
| remove local neighborhood summarization step (simple partitioner) (ablation) | Placeto · nmt dataset | placement runtime: 13% slower | Hurt |
Closest papers by meaning
- A Structure-Aware Framework for Learning Device Placements on Computation Graphs (NeurIPS 2024 · similarity 0.84)
- Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer (NeurIPS 2024 · similarity 0.82)
- Chip Placement with Diffusion Models (ICML 2025 · similarity 0.81)
- Reinforcement Learning within Tree Search for Fast Macro Placement (ICML 2024 · similarity 0.81)
- DistRL: An Asynchronous Distributed Reinforcement Learning Framework for On-Device Control Agent (ICLR 2025 · similarity 0.80)
- A Hierarchical Adaptive Multi-Task Reinforcement Learning Framework for Multiplier Circuit Design (ICML 2024 · similarity 0.80)
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.
