Research map / Architectures and efficiency / RNNs and normalization
A Deep Learning Dataloader with Shared Data Preparation
Jian Xie, Jingwei Xu, Guochang Wang, Yuan Yao, Zenan Li, Chun Cao, Hanghang Tong
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 PyTorch default Baseline dataloader with JOADER | ResNet18 training · ImageNet, 6 parallel jobs | epoch time (min): 67.85 -> 37.55 | Improved |
| Replace PyTorch default Baseline dataloader with JOADER | ResNet18 training · ImageNet, 6 parallel jobs | CPU utilization: over 40% less CPU utilization | Improved |
| JOADER instead of CoorDL when jobs start simultaneously | ResNet18 training · ImageNet, 6 parallel jobs | epoch time (min): 37.98 -> 37.55 | No clear effect |
| JOADER instead of CoorDL for a single job | ResNet18 training · ImageNet, 1 job | epoch time (min): 32.82 -> 32.70 | No clear effect |
| JOADER instead of CoorDL with 4 and 5 jobs | ResNet18 training · ImageNet, 4-5 jobs | epoch time (min): 34.76 -> 36.32 | Hurt |
| JOADER instead of CoorDL, models with different speeds | ResNet18 within five-ResNet mix · ImageNet, ResNet18-152 | speed: 4x faster | Improved |
| Larger models in the mix | JOADER · ImageNet, ResNet101/152 | benefit | Hurt |
| JOADER vs Baseline when jobs arrive at different moments | ResNet18 training · ImageNet, four jobs | training time: saves 17.1% | Improved |
| JOADER vs Baseline when jobs arrive at different moments | ResNet18 training · ImageNet, four jobs | CPU utilization: saves 8.7% | Improved |
| Increase dataset overlap (intersection size) between two jobs | JOADER · ImageNet resampled | CPU utilization: 60% for same dataset vs different datasets | Improved |
| Half of datasets in intersection for 2 jobs | JOADER vs Baseline · ImageNet resampled | CPU utilization: 10% reduction | Improved |
| Remove DSA or RefCnt (ablation) (ablation) | JOADER · ImageNet | training efficiency | Improved |
Closest papers by meaning
- SAPipe: Staleness-Aware Pipeline for Data Parallel DNN Training (NeurIPS 2022 · similarity 0.83)
- Nimble: Lightweight and Parallel GPU Task Scheduling for Deep Learning (NeurIPS 2020 · similarity 0.83)
- Scale Efficient Training for Large Datasets (CVPR 2025 · similarity 0.81)
- Auto-reconfiguration for Latency Minimization in CPU-based DNN Serving (ICML 2025 · similarity 0.81)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.81)
- SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget (ICLR 2025 · 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.
