Yuchen Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Yuchen Liu has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Shu Yin
School of Information Science and Technology, ShanghaiTech University, Shanghai, China
3 shared publications - Guancheng Li
ShanghaiTech University, Shanghai, China
1 shared publication - Junchao Wang
China Meteorological Administration, Basin Heavy Rainfall Key Lab/Hubei Key Lab for HV. Rain Mtrg. & WRNG., Wuhan, Hubei, China
1 shared publication - Kaiyuan Xu
School of Information Science and Technology, ShanghaiTech University, Shanghai, Shanghai, China
1 shared publication - Tianyuan Wu
School of Information Science and Technology, ShanghaiTech University, Shanghai, China
1 shared publication - Tianyuan Wu
School of Info. Sci. & Tech., ShanghaiTech University, Shanghai, China
1 shared publication
Latest Publications
- Open Access
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 823-832
PINCH: Predictive Importance-Sampling-Informed Cache for I/O-Bound DNN Training
DNN training on large datasets is often bottlenecked by data I/O rather than GPU computation. Shuffled SGD destroys temporal locality, rendering conventional caching and prefetching ineffective. Two classes of system address this: clairvoyant caches that exploit the deterministic access trace of seeded shuffling to plan storage proactively, and importance-aware caches that direct resources toward high-value samples. No system combines both because they impose contradictory requirements: clairvoyant planning needs a fixed training set for trace predictability, while importance sampling needs a dynamic subset that adapts to evolving sample scores. We propose PINCH, a cache middleware that decouples the two at the granularity of a planning horizon. At each horizon boundary PINCH re-selects the training subset using importance and stability signals; within the horizon the subset is fixed and the trace is deterministic, enabling planned cache retention and bounded prefetch. Evaluation shows that PINCH achieves up to 12.2 × training speedup over a no-cache baseline, outperforming both clairvoyant-only and importance-only baselines at every operating point.
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 823-832
Open AccessPINCH: Predictive Importance-Sampling-Informed Cache for I/O-Bound DNN Training
DNN training on large datasets is often bottlenecked by data I/O rather than GPU computation. Shuffled SGD destroys temporal locality, rendering conventional caching and prefetching ineffective. Two classes of system address this: clairvoyant caches that exploit the deterministic access trace of seeded shuffling to plan storage proactively, and importance-aware caches that direct resources toward high-value samples. No system combines both because they impose contradictory requirements: clairvoyant planning needs a fixed training set for trace predictability, while importance sampling needs a dynamic subset that adapts to evolving sample scores. We propose PINCH, a cache middleware that decouples the two at the granularity of a planning horizon. At each horizon boundary PINCH re-selects the training subset using importance and stability signals; within the horizon the subset is fixed and the trace is deterministic, enabling planned cache retention and bounded prefetch. Evaluation shows that PINCH achieves up to 12.2 × training speedup over a no-cache baseline, outperforming both clairvoyant-only and importance-only baselines at every operating point.
- Open Access
- 20 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 883-897
A Data Optimizer for Region-Aware Self-describing Files in Scientific Computing
Acquiring data from scientific simulations for analytical purposes is inherently challenging due to the complex and irregularly shaped regions within which the data resides, particularly when using self-describing data formats. The process of region-based data distillation becomes even more arduous when employing persistent memory or parallel file systems. To tackle this challenge, we introduce RASTER (Region-Aware Self-describing daTa optimizER), a lightweight middleware designed for region-aware data preprocessing. RASTER dynamically reorganizes data into variable groups based on regional identifiers during runtime, thereby eliminating the need for sequential searches to locate the required data. We have developed a prototype of RASTER and successfully integrated it into three distinct computing environments: a single-node server equipped with Intel® Optane™ DC persistent memory, the Huawei® OceanStor cloud storage platform, and the Sunway TaihuLight supercomputer. We then conducted a thorough evaluation of the RASTER prototype on the latter two platforms using a real-world scientific application, CESM (Community Earth System Model). Our experimental results demonstrate that RASTER enhances data acquisition performance by up to 2.83× and achieves a 2.36× speedup over conventional netCDF and the state-of-the-art ADIOS2. Additionally, RASTER significantly reduces memory usage by up to 400%, showcasing its scalability potential.
- 20 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 883-897
Open AccessA Data Optimizer for Region-Aware Self-describing Files in Scientific Computing
Acquiring data from scientific simulations for analytical purposes is inherently challenging due to the complex and irregularly shaped regions within which the data resides, particularly when using self-describing data formats. The process of region-based data distillation becomes even more arduous when employing persistent memory or parallel file systems. To tackle this challenge, we introduce RASTER (Region-Aware Self-describing daTa optimizER), a lightweight middleware designed for region-aware data preprocessing. RASTER dynamically reorganizes data into variable groups based on regional identifiers during runtime, thereby eliminating the need for sequential searches to locate the required data. We have developed a prototype of RASTER and successfully integrated it into three distinct computing environments: a single-node server equipped with Intel® Optane™ DC persistent memory, the Huawei® OceanStor cloud storage platform, and the Sunway TaihuLight supercomputer. We then conducted a thorough evaluation of the RASTER prototype on the latter two platforms using a real-world scientific application, CESM (Community Earth System Model). Our experimental results demonstrate that RASTER enhances data acquisition performance by up to 2.83× and achieves a 2.36× speedup over conventional netCDF and the state-of-the-art ADIOS2. Additionally, RASTER significantly reduces memory usage by up to 400%, showcasing its scalability potential.
- 11 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 33 (9) , 2047-2049
Reproducibility: Performance Evaluation of MemXCT on Azure CycleCloud Platform
Memory-Centric X-ray Computational Tomography(CT) is an iterative reconstruction technique that trades compute simplifications with higher memory accesses. MemXCT implements a sparse matrix-vector multiplication(SpMV) with multi-stage buffering and two-level pseudo-Hilbert ordering for optimization. Motivated by the need to validate conclusions from previous work, we reproduce the numerical results, the algorithm’s performance, and the scaling behavior of the algorithms as the number of MPI processes increases on Azure. Digital artifacts from these experiments are available at: 10.5281/zenodo.5598108
- 11 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 33 (9) , 2047-2049
Reproducibility: Performance Evaluation of MemXCT on Azure CycleCloud Platform
Memory-Centric X-ray Computational Tomography(CT) is an iterative reconstruction technique that trades compute simplifications with higher memory accesses. MemXCT implements a sparse matrix-vector multiplication(SpMV) with multi-stage buffering and two-level pseudo-Hilbert ordering for optimization. Motivated by the need to validate conclusions from previous work, we reproduce the numerical results, the algorithm’s performance, and the scaling behavior of the algorithms as the number of MPI processes increases on Azure. Digital artifacts from these experiments are available at: 10.5281/zenodo.5598108