Research map / Reinforcement learning / Visual and representation RL
Inductive-bias-driven Reinforcement Learning For Efficient Schedules in Heterogeneous Clusters
Subho Banerjee, Saurabh Jha, Zbigniew Kalbarczyk, Ravishankar Iyer
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 |
|---|---|---|---|
| Encode architectural resource dependencies in a Bayesian belief model instead of using a model-free RNN | Symphony scheduling · Scheduling workloads on heterogeneous clusters | sample efficiency: ~2.2× worse than Symphony | Improved |
| Use sampling-based Bayesian gradient computation instead of full belief-propagation inference (ablation) | Symphony training · Training to reach 30% mean loss relative to oracle | sample efficiency: at least 4.3× | Improved |
| Use task batching rather than scheduling single tasks (ablation) | Symphony · Scheduler tail latency | tail latency: 12× | Improved |
| Set task batch size to 128 (ablation) | Symphony · Task batching | oracle-normalized workload latency: 128 tasks per batch | Improved |
| Use random task assignment rather than Symphony's learned scheduling policy | Scheduling workloads on heterogeneous clusters · Scheduling latency and workload latency | workload latency: 3.6× lower median latency | Mixed |
| Account for system topology and low-level resource contention in scheduling | Symphony · Three workloads, 500 executions | oracle-normalized kernel runtime: 32% | Improved |
Closest papers by meaning
- Scheduling of Time-Varying Workloads Using Reinforcement Learning (AAAI 2021 · similarity 0.87)
- Towards Generalizable Multi-Policy Optimization with Self-Evolution for Job Scheduling (NeurIPS 2025 · similarity 0.86)
- DRIVE: Best Data Scheduling Practices for Reinforcement Learning with Verifiable Reward in Competitive Code Generation (ICML 2026 · similarity 0.85)
- Hard Tasks First: Multi-Task Reinforcement Learning Through Task Scheduling (ICML 2024 · similarity 0.83)
- SustainDC: Benchmarking for Sustainable Data Center Control (NeurIPS 2024 · similarity 0.83)
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.83)
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.
