FusionFlow: Accelerating Data Preprocessing for Machine Learning with CPU-GPU Cooperation

Summary: FusionFlow accelerates dynamic DL augmentation via CPU-GPU co-scheduling, managing shared GPU memory to avoid overflow and interference with training. Adaptive scheduling reallocates resources across heterogeneous tasks, cutting CPU requirements 50–60% while improving throughput. (summarized by gpt-5.6-luna on Jul 24 2026)

Paper ID: 13948
Venue: VLDB
Year: 2024
Pagerank: 5.4544352e-05
Overall Rank: 8,321 | 42.92%
DOI: 10.14778/3636218.3636238

Incoming Non-self Citations Over Time

Authors

1. Taeyoon Kim (Ulsan National Institute of Science and Technology)
2. ChanHo Park (Ulsan National Institute of Science and Technology)
3. Mansur Mukimbekov (Ulsan National Institute of Science and Technology)
4. Heelim Hong (Ulsan National Institute of Science and Technology)
5. Minseok Kim (Ulsan National Institute of Science and Technology)
6. Ze Jin (ByteDance)
7. Changdae Kim (Electronics and Telecommunications Research Institute)
8. Ji-Yong Shin (Northeastern University)
9. Myeongjae Jeon (Ulsan National Institute of Science and Technology)

BibTeX Citation

@article{kim_vldb24,
        title = {{FusionFlow: Accelerating Data Preprocessing for Machine Learning with CPU-GPU Cooperation}},
        author = {Kim, Taeyoon and Park, ChanHo and Mukimbekov, Mansur and Hong, Heelim and Kim, Minseok and Jin, Ze and Kim, Changdae and Shin, Ji-Yong and Jeon, Myeongjae},
        journal = {PVLDB},
        series = {{VLDB} '24},
        volume = {17},
        number = {4},
        pages = {863--876},
        doi = {10.14778/3636218.3636238},
        url = {https://doi.org/10.14778/3636218.3636238},
        year = {2024}
}

Incoming Citations (Sorted by Pagerank)

Showing 2 of 2 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
8,905	TensorSocket: Shared Data Loading for Deep Learning Training	2026	SIGMOD	5.3483178e-05
10,999	cedar: Optimized and Unified Machine Learning Input Data Pipelines	2025	VLDB	5.093636e-05

Outgoing Citations (Sorted by Pagerank)

Showing 5 of 5 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
1,446	Analyzing and Mitigating Data Stalls in DNN Training	2021	VLDB	0.0001076818
2,018	tf.data: A Machine Learning Data Processing Framework	2021	VLDB	9.3001937e-05
3,764	FastFlow: Accelerating Deep Learning Model Training with Smart Offloading of Input Data Pipeline	2023	VLDB	7.1446931e-05
4,956	Heterogeneity-Aware Distributed Machine Learning Training via Partial Reduce	2021	SIGMOD	6.4290135e-05
5,282	GoldMiner: Elastic Scaling of Training Data Pre-Processing Pipelines for Deep Learning	2023	SIGMOD	6.2840712e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	5,075	CGgraph: An Ultra-fast Graph Processing System on Modern Commodity CPU-GPU Co-processor	2024	VLDB
2	10,512	SVFusion: A CPU-GPU Co-Processing Architecture for Large-Scale Real-Time Vector Search	2026	VLDB
3	7,112	Coresets over Multiple Tables for Feature-rich and Data-efficient Machine Learning	2023	VLDB
4	5,282	GoldMiner: Elastic Scaling of Training Data Pre-Processing Pipelines for Deep Learning	2023	SIGMOD
5	5,003	Galvatron: Efficient Transformer Training over Multiple GPUs Using Automatic Parallelism	2023	VLDB
6	8,905	TensorSocket: Shared Data Loading for Deep Learning Training	2026	SIGMOD
7	2,018	tf.data: A Machine Learning Data Processing Framework	2021	VLDB
8	10,219	DFLOP: A Data-driven Framework for Multimodal LLM Training Pipeline Optimization	2026	SIGMOD
9	3,541	Where Is My Training Bottleneck? Hidden Trade-Offs in Deep Learning Preprocessing Pipelines	2022	SIGMOD
10	3,764	FastFlow: Accelerating Deep Learning Model Training with Smart Offloading of Input Data Pipeline	2023	VLDB