Research map / Architectures and efficiency / Mixture of experts
Deft Scheduling of Dynamic Cloud Workflows with Varying Deadlines via Mixture-of-Experts
Ya Shen, Gang Chen, Hui Ma, Mengjie Zhang
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 |
|---|---|---|---|
| Replace the single-path PMM with a deadline-aware mixture of experts | GATES · CADWS dynamic deadlines, L-scale | total cost: 195.65 -> 137.69 | Improved |
| Replace the single-path PMM with MoE | GATES · CADWS dynamic deadlines, S-scale | total cost: 52.95 -> 52.46 | Improved |
| Replace the single-path PMM with MoE | GATES · CADWS dynamic deadlines, M-scale | total cost: 97.76 -> 86.60 | Improved |
| Add graph-adaptive gating instead of linear gating (ablation) | DEFT · CADWS dynamic deadlines, L-scale | total cost: 142.27 -> 137.69 | Improved |
| Add graph-adaptive gating instead of MLP gating (ablation) | DEFT · CADWS dynamic deadlines, M-scale | total cost: 87.34 -> 86.60 | Improved |
| Use a deeper MLP priority mapping module instead of MoE (ablation) | GATES · CADWS dynamic deadlines, L-scale | total cost: 195.65 -> 194.77 | No clear effect |
| Use 4 experts instead of 2 (ablation) | DEFT · CADWS ablation, M-scale | total cost: 32.09 -> 28.32 | Improved |
| Use 4 experts instead of 8 (ablation) | DEFT · CADWS ablation, M-scale | total cost: 32.67 -> 28.32 | Improved |
| Use Top-1 routing instead of Top-2 (ablation) | DEFT with 4 experts · CADWS ablation, M-scale | total cost: 31.48 -> 28.32 | Improved |
| Use Top-1 routing instead of Top-2 (ablation) | DEFT with 2 experts · CADWS ablation, S-scale | total cost: 20.56 -> 19.80 | Improved |
| Add MoE with graph-adaptive gating to GATES | GATES · CADWS inference overhead (overall) | seconds/step: 0.2159 -> 0.2218 | Hurt |
| Use DEFT instead of GATES under fixed deadline | GATES · CADWS gamma=1.25, L-scale | total cost: 105.33 -> 99.16 | Improved |
Closest papers by meaning
- FlowMoE: A Scalable Pipeline Scheduling Framework for Distributed Mixture-of-Experts Training (NeurIPS 2025 · similarity 0.84)
- Beyond Static Pipelines: Learning Dynamic Workflows for Text-to-SQL (ICML 2026 · similarity 0.83)
- HybridFlow: Resource-Adaptive Subtask Routing for Efficient Edge-Cloud LLM Inference (ICML 2026 · similarity 0.83)
- DCcluster-Opt: Benchmarking Dynamic Multi-Objective Optimization for Geo-Distributed Data Center Workloads (NeurIPS 2025 · similarity 0.83)
- Optimize Once, Execute Fast: Latency-Aware Multi-Agent Workflow Learning for Recurrent Queries (NeurIPS 2026 · similarity 0.82)
- DynaSchedBench: Calibrated Dynamic Scheduling Benchmarks and Observability Paradox in LLM-based Scheduling Agents (ICML 2026 · similarity 0.82)
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.
