Ce Yu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ce Yu has a total of 190 co-authors. Here are the top co-authors by number of shared publications:
Jian Xiao
College of Intelligence and Computing, Tianjin University , No. 135 Yaguan Road, Haihe Education Park, Tianjin 300350, China
17 shared publications- Shanjiang Tang
College of Intelligence and Computing, Tianjin University, Tianjin, China
13 shared publications Chao Sun
College of Intelligence and Computing Tianjin University No.135 Yaguan Road Haihe Education Park 300350 Tianjin China
11 shared publications- Jian Xiao
College of Intelligence and Computing, Tianjin University, Tianjin, China
11 shared publications Jizhou Sun
Tianjin University
11 shared publications- Chenzhou Cui
National Astronomical Observatories, Chinese Academy of Sciences
9 shared publications
Latest Publications
- Open Access
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1577-1592
Storage Evolution for Exascale Supercomputers: Lessons from Sunway TaihuLight to OceanLight
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1577-1592
Open AccessStorage Evolution for Exascale Supercomputers: Lessons from Sunway TaihuLight to OceanLight
- 27 October 2026
- Published by Wiley in WIREs Data Mining and Knowledge Discovery
- Vol. 16 (4)
Scheduling Techniques in Astronomical Observation: Survey, Taxonomy, Challenges, and Future Trends
Astronomical observation scheduling becomes a prominent issue due to the urgent need for astronomers to conduct a comprehensive exploration of the dynamic universe. In recent decades, an increase in survey area and spatio‐temporal monitoring frequency is raising the complexity of scheduling decisions. Inefficient scheduling technologies face issues such as inefficient and inflexible decision‐making, overutilized resources, and imbalanced resource allocation, which are not conducive to scientific output. Astronomical observation facilities are expensive and have a limited life cycle, so the improvement of observation efficiency is crucial. A variety of time‐varying constraints, such as scientific objectives, observation resource performance, and astronomical observation conditions, pose great challenges to this problem. Hence, proper strategies for task scheduling and telescope coordination are indispensable. A comprehensive taxonomic review of the current state‐of‐the‐art scheduling techniques is presented using exhaustive evaluation criteria. An extensive list of scheduling objectives, along with their associated evaluation metrics, has been considered in this study. A rigorous evaluation of existing literature based on key features is conducted, and their limitations are highlighted. Many factors and challenges in advancing astronomical observation scheduling for next‐generation automatic astronomical applications are also identified, and several potential recommendations are provided. This article is categorized under: Technologies > Artificial Intelligence Technologies > Machine Learning Application Areas > Science and Technology
- 27 October 2026
- Published by Wiley in WIREs Data Mining and Knowledge Discovery
- Vol. 16 (4)
Scheduling Techniques in Astronomical Observation: Survey, Taxonomy, Challenges, and Future Trends
Astronomical observation scheduling becomes a prominent issue due to the urgent need for astronomers to conduct a comprehensive exploration of the dynamic universe. In recent decades, an increase in survey area and spatio‐temporal monitoring frequency is raising the complexity of scheduling decisions. Inefficient scheduling technologies face issues such as inefficient and inflexible decision‐making, overutilized resources, and imbalanced resource allocation, which are not conducive to scientific output. Astronomical observation facilities are expensive and have a limited life cycle, so the improvement of observation efficiency is crucial. A variety of time‐varying constraints, such as scientific objectives, observation resource performance, and astronomical observation conditions, pose great challenges to this problem. Hence, proper strategies for task scheduling and telescope coordination are indispensable. A comprehensive taxonomic review of the current state‐of‐the‐art scheduling techniques is presented using exhaustive evaluation criteria. An extensive list of scheduling objectives, along with their associated evaluation metrics, has been considered in this study. A rigorous evaluation of existing literature based on key features is conducted, and their limitations are highlighted. Many factors and challenges in advancing astronomical observation scheduling for next‐generation automatic astronomical applications are also identified, and several potential recommendations are provided. This article is categorized under: Technologies > Artificial Intelligence Technologies > Machine Learning Application Areas > Science and Technology
- Open Access
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 100-110
ZSWP: Rollback-Free Spherical ε-Join for Scalable Astronomical Cross-Matching on Heterogeneous Supercomputer Architectures
Spherical ϵ -join tasks, including astronomical cross-matching, satellite conjunction assessment, and global geospatial fusion, require exact pairwise distance checks on the unit sphere at billion-object scales. While modern heterogeneous supercomputers offer substantial compute throughput, efficient data organization and coordination across deep memory hierarchies remain a critical bottleneck. Traditional spherical partitioning methods, such as the Zones algorithm, inherently suffer from window rollback: a non-monotonic data-access pattern arising from the latitude-dependent window radius. On current accelerators, this irregularity undermines cache reuse, obstructs hardware prefetching, and degrades pipeline efficiency. To address this, we propose Zones-based Sliding-Window Pruning (ZSWP), a lightweight, linear-time reorganization scheme. ZSWP partitions source stripes into sub-stripes with strictly monotonic window advancement, eliminating rollback by construction while preserving exact join semantics. The resulting forward-only access pattern enables a hierarchy-aware pipeline that maps the continuous candidate data stream directly to the heterogeneous memory layers, including DDR, global shared memory (GSM), and array memory (AM). This hardware-software co-design enables segmented data streaming, safe cache replacement, and double-buffered vector execution. Evaluated on real astronomical datasets, our method guarantees 100% scientific correctness and achieves near-ideal strong scalability (32—35 × speedup) and stable weak scalability across 4—128 heterogeneous nodes.
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 100-110
Open AccessZSWP: Rollback-Free Spherical ε-Join for Scalable Astronomical Cross-Matching on Heterogeneous Supercomputer Architectures
Spherical ϵ -join tasks, including astronomical cross-matching, satellite conjunction assessment, and global geospatial fusion, require exact pairwise distance checks on the unit sphere at billion-object scales. While modern heterogeneous supercomputers offer substantial compute throughput, efficient data organization and coordination across deep memory hierarchies remain a critical bottleneck. Traditional spherical partitioning methods, such as the Zones algorithm, inherently suffer from window rollback: a non-monotonic data-access pattern arising from the latitude-dependent window radius. On current accelerators, this irregularity undermines cache reuse, obstructs hardware prefetching, and degrades pipeline efficiency. To address this, we propose Zones-based Sliding-Window Pruning (ZSWP), a lightweight, linear-time reorganization scheme. ZSWP partitions source stripes into sub-stripes with strictly monotonic window advancement, eliminating rollback by construction while preserving exact join semantics. The resulting forward-only access pattern enables a hierarchy-aware pipeline that maps the continuous candidate data stream directly to the heterogeneous memory layers, including DDR, global shared memory (GSM), and array memory (AM). This hardware-software co-design enables segmented data streaming, safe cache replacement, and double-buffered vector execution. Evaluated on real astronomical datasets, our method guarantees 100% scientific correctness and achieves near-ideal strong scalability (32—35 × speedup) and stable weak scalability across 4—128 heterogeneous nodes.
- Open Access
- 7 August 2026
- Published by American Astronomical Society in The Astrophysical Journal
- Vol. 1007 (1) , 82
Reframing Variable Star Classification as Structured Sequence Modeling with Large Language Models
Variable stars play a crucial role in advancing knowledge of stellar evolution, Galactic structure, and cosmological distance scales. The advent of massive datasets from modern time-domain surveys (e.g., Zwicky Transient Facility, All-Sky Automated Survey for Supernovae, or Gaia) has made the classification of these stars increasingly complex. Traditional methods depend extensively on manual feature engineering, a process that is both labor-intensive and difficult to scale, thereby impeding timely scientific progress. To address these limitations, we propose a novel classification framework that treats photometric light curves as structured sequences and fine-tunes a pretrained large language model (LLM) for variable star classification with only a small amount of labeled data. We design a unified tokenization scheme that integrates both time-series measurements and physical parameters (period, parallax, and color index) into a single textual prompt, thereby streamlining traditional feature engineering. When evaluated on a multisurvey dataset comprising 19,580 labeled variable stars across nine classes, the optimal model achieves an overall accuracy of 0.98 and a weighted F 1 score of 0.98. We also verify the complementary contributions of each physical parameter through ablation studies. Our study indicates that LLM-based sequence modeling constitutes an accurate, scalable, and flexible alternative to conventional methods, with strong potential for integration into future large-scale classification of variable stars.
- 7 August 2026
- Published by American Astronomical Society in The Astrophysical Journal
- Vol. 1007 (1) , 82
Open AccessReframing Variable Star Classification as Structured Sequence Modeling with Large Language Models
Variable stars play a crucial role in advancing knowledge of stellar evolution, Galactic structure, and cosmological distance scales. The advent of massive datasets from modern time-domain surveys (e.g., Zwicky Transient Facility, All-Sky Automated Survey for Supernovae, or Gaia) has made the classification of these stars increasingly complex. Traditional methods depend extensively on manual feature engineering, a process that is both labor-intensive and difficult to scale, thereby impeding timely scientific progress. To address these limitations, we propose a novel classification framework that treats photometric light curves as structured sequences and fine-tunes a pretrained large language model (LLM) for variable star classification with only a small amount of labeled data. We design a unified tokenization scheme that integrates both time-series measurements and physical parameters (period, parallax, and color index) into a single textual prompt, thereby streamlining traditional feature engineering. When evaluated on a multisurvey dataset comprising 19,580 labeled variable stars across nine classes, the optimal model achieves an overall accuracy of 0.98 and a weighted F 1 score of 0.98. We also verify the complementary contributions of each physical parameter through ablation studies. Our study indicates that LLM-based sequence modeling constitutes an accurate, scalable, and flexible alternative to conventional methods, with strong potential for integration into future large-scale classification of variable stars.
- Open Access
- 28 July 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 285 (2) , 57
TDL ight : A Framework for Incremental Light-curve Management and Continuous Classification Optimization
Time-series observations of celestial objects are fundamental to studying variable and transient phenomena. As time-domain surveys continue to scale up, timely automated classification is essential for rapid follow-up. Many existing approaches treat storage and analysis as separate stages, archiving data first and analyzing it later. Yet each source is a continuously growing stream, and early observations often contain sufficient information to assign scientifically useful labels before the full light curve is available. Motivated by this, we present TDL ight , a framework for light-curve management and trigger-based classification over incremental data streams. Built on the time-series database TD ENgine , the system loads the Gaia DR2 archive (∼4.8 × 10 7 rows) in 33 s and supports low-latency retrieval. As new observations accumulate, TDL ight automatically triggers a lightweight LightGBM-based classification pipeline that requires no GPU. When prediction confidence exceeds a predefined threshold, these early results are written back to the database, providing reliable classifications before offline analysis is complete. The deployed system also supports manual label confirmation and optional incremental training. Evaluated on the LEAVES dataset (∼1.1 million sources spanning 10 classes), the system achieves an overall accuracy of 93.6% and a weighted F1 score of 93.7%. Furthermore, it identifies 560 high-confidence catalog-disagreement candidates for follow-up verification. Notably, this includes correcting 150 RRc variables erroneously labeled as W Ursae Majoris binaries due to inherited period-doubling artifacts. TDL ight and the related data products are publicly available.
- 28 July 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 285 (2) , 57
Open AccessTDL ight : A Framework for Incremental Light-curve Management and Continuous Classification Optimization
Time-series observations of celestial objects are fundamental to studying variable and transient phenomena. As time-domain surveys continue to scale up, timely automated classification is essential for rapid follow-up. Many existing approaches treat storage and analysis as separate stages, archiving data first and analyzing it later. Yet each source is a continuously growing stream, and early observations often contain sufficient information to assign scientifically useful labels before the full light curve is available. Motivated by this, we present TDL ight , a framework for light-curve management and trigger-based classification over incremental data streams. Built on the time-series database TD ENgine , the system loads the Gaia DR2 archive (∼4.8 × 10 7 rows) in 33 s and supports low-latency retrieval. As new observations accumulate, TDL ight automatically triggers a lightweight LightGBM-based classification pipeline that requires no GPU. When prediction confidence exceeds a predefined threshold, these early results are written back to the database, providing reliable classifications before offline analysis is complete. The deployed system also supports manual label confirmation and optional incremental training. Evaluated on the LEAVES dataset (∼1.1 million sources spanning 10 classes), the system achieves an overall accuracy of 93.6% and a weighted F1 score of 93.7%. Furthermore, it identifies 560 high-confidence catalog-disagreement candidates for follow-up verification. Notably, this includes correcting 150 RRc variables erroneously labeled as W Ursae Majoris binaries due to inherited period-doubling artifacts. TDL ight and the related data products are publicly available.
- Open Access
TDLight: A Framework for Incremental Light Curve Management and Smart Classification
With the exponential growth of time-domain surveys, the volume of light curves has increased rapidly. However, many survey projects, such as Gaia, still rely on offline batch-processing workflows in which data are calibrated, merged, and released only after an observing phase is completed. This latency delays scientific analysis and causes many high-value transient events to be buried in archival data, missing the window for timely follow-up. While existing alert brokers handle heterogeneous data streams, it remains difficult to deploy a unified framework that combines high-performance incremental storage with real-time classification on local infrastructure. To address this challenge, we propose TDLight, a scalable system that adapts the time-series database TDengine (a high-performance IoT database) for astronomical data using a one-table-per-source schema. This architecture supports high-throughput ingestion, achieving 954,000 rows s^-1 for archived data and 541,000 rows s^-1 for incremental streams, while Hierarchical Equal Area isoLatitude Pixelization (HEALPix) indexing enables efficient cone-search queries. Building on this storage layer, we integrate the pre-trained hierarchical Random Forest classifier from the LEAVES framework to construct an incremental classification pipeline. Using the LEAVES dataset, we simulate data accumulation and evaluate a trigger-based strategy that performs early classification at specific observational milestones. In addition, by monitoring the evolution of classification probabilities, the system identifies "high-value candidates" -- sources that show high early confidence but later undergo significant label shifts. TDLight is released as an open-source Dockerized environment, providing a deployable infrastructure for next-generation time-domain surveys.
Open AccessTDLight: A Framework for Incremental Light Curve Management and Smart Classification
With the exponential growth of time-domain surveys, the volume of light curves has increased rapidly. However, many survey projects, such as Gaia, still rely on offline batch-processing workflows in which data are calibrated, merged, and released only after an observing phase is completed. This latency delays scientific analysis and causes many high-value transient events to be buried in archival data, missing the window for timely follow-up. While existing alert brokers handle heterogeneous data streams, it remains difficult to deploy a unified framework that combines high-performance incremental storage with real-time classification on local infrastructure. To address this challenge, we propose TDLight, a scalable system that adapts the time-series database TDengine (a high-performance IoT database) for astronomical data using a one-table-per-source schema. This architecture supports high-throughput ingestion, achieving 954,000 rows s^-1 for archived data and 541,000 rows s^-1 for incremental streams, while Hierarchical Equal Area isoLatitude Pixelization (HEALPix) indexing enables efficient cone-search queries. Building on this storage layer, we integrate the pre-trained hierarchical Random Forest classifier from the LEAVES framework to construct an incremental classification pipeline. Using the LEAVES dataset, we simulate data accumulation and evaluate a trigger-based strategy that performs early classification at specific observational milestones. In addition, by monitoring the evolution of classification probabilities, the system identifies "high-value candidates" -- sources that show high early confidence but later undergo significant label shifts. TDLight is released as an open-source Dockerized environment, providing a deployable infrastructure for next-generation time-domain surveys.
- 21 September 2025
- Published by Springer Nature in Journal of Visualization
- Vol. 28 (6) , 1103-1117
A light multimodal neural network for tropical cyclone path prediction with single time step
- 21 September 2025
- Published by Springer Nature in Journal of Visualization
- Vol. 28 (6) , 1103-1117
A light multimodal neural network for tropical cyclone path prediction with single time step
- Open Access
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 11-21
MixLoRA: An Efficient Multi-Tenant Framework for Concurrently Serving Diverse LoRA Models in Large Language Models
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 11-21
Open AccessMixLoRA: An Efficient Multi-Tenant Framework for Concurrently Serving Diverse LoRA Models in Large Language Models
- 21 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 36 (10) , 2073-2088
Task Scheduling in Geo-Distributed Computing: A Survey
Geo-distributed computing, a paradigm that assigns computational tasks to globally distributed nodes, has emerged as a promising approach in cloud computing, edge computing, cloud-edge computing, and supercomputer computing (SC). It enables low-latency services, ensures data locality, and handles large-scale applications. As global computing capacity and task demands increase rapidly, scheduling tasks for efficient execution in geo-distributed computing systems has become an increasingly critical research challenge. It arises from the inherent characteristics of geographic distribution, including heterogeneous network conditions, region-specific resource pricing, and varying computational capabilities across locations. Researchers have developed diverse task scheduling methods tailored to geo-distributed scenarios, aiming to achieve objectives such as performance enhancement, fairness assurance, and fault-tolerance improvement. This survey provides a comprehensive and systematic review of task scheduling techniques across four major distributed computing environments, with an in-depth analysis of these approaches based on their core scheduling objectives. Through our analysis, we identify key research challenges and outline promising directions for advancing task scheduling in geo-distributed computing.
- 21 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 36 (10) , 2073-2088
Task Scheduling in Geo-Distributed Computing: A Survey
Geo-distributed computing, a paradigm that assigns computational tasks to globally distributed nodes, has emerged as a promising approach in cloud computing, edge computing, cloud-edge computing, and supercomputer computing (SC). It enables low-latency services, ensures data locality, and handles large-scale applications. As global computing capacity and task demands increase rapidly, scheduling tasks for efficient execution in geo-distributed computing systems has become an increasingly critical research challenge. It arises from the inherent characteristics of geographic distribution, including heterogeneous network conditions, region-specific resource pricing, and varying computational capabilities across locations. Researchers have developed diverse task scheduling methods tailored to geo-distributed scenarios, aiming to achieve objectives such as performance enhancement, fairness assurance, and fault-tolerance improvement. This survey provides a comprehensive and systematic review of task scheduling techniques across four major distributed computing environments, with an in-depth analysis of these approaches based on their core scheduling objectives. Through our analysis, we identify key research challenges and outline promising directions for advancing task scheduling in geo-distributed computing.
- 1 March 2025
- Published by IOP Publishing in Publications of the Astronomical Society of the Pacific
- Vol. 137 (3) , 034509
SPGrid: A Straightforward Parallel Gridding Approach for Single-dish Radio Telescope
The gridding operation, resampling irregularly distributed raw data onto a regular grid map, is the most data-intensive and compute-intensive step in radio astronomical data processing pipelines. Therefore, in order to tackle the vast amount of data generated by modern large radio telescopes, many parallel convolution-based gridding algorithms are developed on multi-core CPU platforms or heterogeneous computing platforms to take full advantage of the processors. Most of these algorithms adopted gather mode, which relies on a fast lookup scheme to avoid write conflict and improve throughput. However, practical applications have shown that constructing a lookup table incurs substantial costs and usually becomes the bottleneck of the gather mode when handling large data sets. So we employ a straightforward parallel gridding approach, that is the scatter mode, to develop the SPGrid code. Several optimization strategies are proposed for CPU and GPU architecture separately. The code improves throughput further while avoiding the heavy overhead of constructing a lookup table. Compared with the representative gather-based gridding code such as Cygrid and HCGrid, the scatter-based SPGrid achieves a maxiumum 4.1× and 3.7× speedup on CPU and GPU platform, respectively.
- 1 March 2025
- Published by IOP Publishing in Publications of the Astronomical Society of the Pacific
- Vol. 137 (3) , 034509
SPGrid: A Straightforward Parallel Gridding Approach for Single-dish Radio Telescope
The gridding operation, resampling irregularly distributed raw data onto a regular grid map, is the most data-intensive and compute-intensive step in radio astronomical data processing pipelines. Therefore, in order to tackle the vast amount of data generated by modern large radio telescopes, many parallel convolution-based gridding algorithms are developed on multi-core CPU platforms or heterogeneous computing platforms to take full advantage of the processors. Most of these algorithms adopted gather mode, which relies on a fast lookup scheme to avoid write conflict and improve throughput. However, practical applications have shown that constructing a lookup table incurs substantial costs and usually becomes the bottleneck of the gather mode when handling large data sets. So we employ a straightforward parallel gridding approach, that is the scatter mode, to develop the SPGrid code. Several optimization strategies are proposed for CPU and GPU architecture separately. The code improves throughput further while avoiding the heavy overhead of constructing a lookup table. Compared with the representative gather-based gridding code such as Cygrid and HCGrid, the scatter-based SPGrid achieves a maxiumum 4.1× and 3.7× speedup on CPU and GPU platform, respectively.