Research map / Architectures and efficiency / RNNs and normalization
Multi-Task Learning for Straggler Avoiding Predictive Job Scheduling
Neeraja J. Yadwadkar, Bharath Hariharan, Joseph E. Gonzalez, Randy Katz
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 |
|---|---|---|---|
| add node-specific and workload-specific weights to a global weight vector (ablation) | f0 · Straggler prediction (50% training data) | prediction accuracy (%): 68.5 -> 75.3 | Improved |
| use node-specific weights instead of a single global vector (ablation) | f0 · Straggler prediction (50% training data) | prediction accuracy (%): 68.5 -> 74.1 | Improved |
| use workload-specific weights instead of a single global vector (ablation) | f0 · Straggler prediction (50% training data) | prediction accuracy (%): 68.5 -> 70.4 | Improved |
| add workload weights to node weights (f0,n,l vs fn) (ablation) | fn · Straggler prediction (50% training data) | prediction accuracy (%): 74.1 -> 75.3 | Improved |
| replace Wrangler per-node-workload models with MTL formulation f0,t,l | Wrangler · Straggler prediction (50% training data) | prediction accuracy (%): 70.1 -> 77.2 | Improved |
| replace Wrangler with MTL f0,t at small training data | Wrangler · Straggler prediction (10% training data) | prediction accuracy (%): 63.9 -> 73.1 | Improved |
| add per node-workload weights to f0,n,l (f0,t,l vs f0,n,l) (ablation) | f0,n,l · Straggler prediction (50% training data) | prediction accuracy (%): 75.3 -> 77.2 | Improved |
| select partitions automatically with mixed l1/l2 norm (ablation) | f0,n,l · Straggler prediction (50% training data) | prediction accuracy (%): 75.3 -> 74.4 | No clear effect |
| select groups automatically with mixed l1/l2 norm (ablation) | f0,n,l · Straggler prediction (50% training data) | prediction accuracy (%): 75.3 -> 73.8 | No clear effect |
| select feature blocks automatically (ffs1, ffs2) (ablation) | f0,n,l · Straggler prediction (test set) | prediction accuracy (%): 75.3 -> 74.9 | No clear effect |
| remove per node-workload weights so unseen node-workload pairs use node and workload weights (f0,n,l vs f0,t) (ablation) | f0,t · Unseen node-workload pair (20 runs) | prediction accuracy (%): 13 out of 20 | Improved |
| use MTL f0,n,l instead of Wrangler | Wrangler · 99th percentile job completion time, FB2009 | improvement over Wrangler (%): 57.8% | Improved |
Closest papers by meaning
- ATLAS: Alibaba Dataset and Benchmark for Learning-Augmented Scheduling (ICLR 2026 · similarity 0.84)
- Revisiting Fairness in Multitask Learning: A Performance-Driven Approach for Variance Reduction (CVPR 2025 · similarity 0.82)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.82)
- Quantifying Task Priority for Multi-Task Optimization (CVPR 2024 · similarity 0.81)
- DMTG: One-Shot Differentiable Multi-Task Grouping (ICML 2024 · similarity 0.81)
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.81)
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.
