Jiansong Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jiansong Li has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Lei Liu
Chinese Academy of Sciences
4 shared publications - Guangli Li
Institute of Computing Technology, CAS
3 shared publications - Linze Li
CTG Wuhan Science and Technology Innovation Park, China Three Gorges Corporation, Wuhan, China
3 shared publications - Peng Zhao
Huawei 2012 Lab, Beijing, China
3 shared publications - Wei Cao
Shenyang Aerospace University (China)
3 shared publications - Xiao Dong
Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China
3 shared publications
Latest Publications
- Open Access
- 10 February 2026
- Published by MDPI AG in Sustainability
- Vol. 18 (4) , 1806
Regional Ecological Security Assessment and Driving Factor Analysis Based on the Innovative Health-Service-Risk-Sensitivity Framework: A Case Study of an Arid Inland River Basin
Under multiple stresses such as an arid climate, water scarcity, and desertification, inland river basins in arid regions represent a typically fragile ecosystem worldwide, and their ecological security faces increasingly complex and severe challenges. To address the limitations of traditional assessment methods characterized by single-perspective approaches, difficulties in quantifying indicators, and lack of a systematic framework for arid basins, this study constructed an innovative Health–Service–Risk–Sensitivity (HSRS) framework. Taking the Tarim River Basin (TRB) as a case study, the validity and necessity of this framework were validated through the Remote Sensing Ecological Index (RSEI) and correlation analysis. Furthermore, the XGBoost–SHAP model was further integrated to identify key threshold responses of multidimensional driving factors within the basin. The findings indicate that the ecological security of the TRB progressively improved, with approximately 11.64% of the area showing significant enhancement. The four most influential driving factors were land use, NDVI, human activity intensity, and soil moisture. Notably, the study identified critical environmental thresholds: when DEM ranged from 1500 to 3000 m and slope from 2° to 30°, constraining effects on the Comprehensive Ecological Security Index (CESI) increased. When annual precipitation exceeded 150 mm, NDVI was greater than 0.35, and soil moisture content exceeded 0.14 m3/m3, the constraint effect was further strengthened. Overall, the integration of the HSRS framework and the XGBoost-SHAP model offers a novel and effective approach for ecological security assessment in arid inland basins. Moreover, this approach has substantial practical implications for achieving precise coordination between regional ecological protection and sustainable development.
- 10 February 2026
- Published by MDPI AG in Sustainability
- Vol. 18 (4) , 1806
Open AccessRegional Ecological Security Assessment and Driving Factor Analysis Based on the Innovative Health-Service-Risk-Sensitivity Framework: A Case Study of an Arid Inland River Basin
Under multiple stresses such as an arid climate, water scarcity, and desertification, inland river basins in arid regions represent a typically fragile ecosystem worldwide, and their ecological security faces increasingly complex and severe challenges. To address the limitations of traditional assessment methods characterized by single-perspective approaches, difficulties in quantifying indicators, and lack of a systematic framework for arid basins, this study constructed an innovative Health–Service–Risk–Sensitivity (HSRS) framework. Taking the Tarim River Basin (TRB) as a case study, the validity and necessity of this framework were validated through the Remote Sensing Ecological Index (RSEI) and correlation analysis. Furthermore, the XGBoost–SHAP model was further integrated to identify key threshold responses of multidimensional driving factors within the basin. The findings indicate that the ecological security of the TRB progressively improved, with approximately 11.64% of the area showing significant enhancement. The four most influential driving factors were land use, NDVI, human activity intensity, and soil moisture. Notably, the study identified critical environmental thresholds: when DEM ranged from 1500 to 3000 m and slope from 2° to 30°, constraining effects on the Comprehensive Ecological Security Index (CESI) increased. When annual precipitation exceeded 150 mm, NDVI was greater than 0.35, and soil moisture content exceeded 0.14 m3/m3, the constraint effect was further strengthened. Overall, the integration of the HSRS framework and the XGBoost-SHAP model offers a novel and effective approach for ecological security assessment in arid inland basins. Moreover, this approach has substantial practical implications for achieving precise coordination between regional ecological protection and sustainable development.
- Open Access
An Improved Quantitative Analysis Method for the Unequal Supply and Demand of Ecosystem Services and Hierarchical Governance Suggestions
Quantifying the unequal supply and demand of ecosystem services (ESs) is a prerequisite for hierarchical ecological governance decisions. However, previous studies have largely overlooked the scale effect of spatially adjacent units and the role of spatial compactness in shaping inequality. To address these research gaps, this study conducted a survey in six counties within the Danjiangkou Basin in China. By adopting a moving window-based local Gini coefficient method, we quantified the inequality in the supply and demand of ESs in this region, and introduced a refined coefficient of variation to measure spatial compactness, analyzing the impact of urbanization on this inequality. The results indicate that the inequality in the supply and demand of ESs in this region is gradually intensifying. However, from a local perspective, the inequality exhibits significant spatial heterogeneity, decreasing gradually from urban centers to suburbs and rural areas, while maintaining strong spatial continuity. Furthermore, we found that urbanization is the primary factor exacerbating this inequality, while compact urban development can mitigate it. The findings of this study can provide practical guidance for cross-county ecological coordination, ecological restoration, and sustainable urban development.
Open AccessAn Improved Quantitative Analysis Method for the Unequal Supply and Demand of Ecosystem Services and Hierarchical Governance Suggestions
Quantifying the unequal supply and demand of ecosystem services (ESs) is a prerequisite for hierarchical ecological governance decisions. However, previous studies have largely overlooked the scale effect of spatially adjacent units and the role of spatial compactness in shaping inequality. To address these research gaps, this study conducted a survey in six counties within the Danjiangkou Basin in China. By adopting a moving window-based local Gini coefficient method, we quantified the inequality in the supply and demand of ESs in this region, and introduced a refined coefficient of variation to measure spatial compactness, analyzing the impact of urbanization on this inequality. The results indicate that the inequality in the supply and demand of ESs in this region is gradually intensifying. However, from a local perspective, the inequality exhibits significant spatial heterogeneity, decreasing gradually from urban centers to suburbs and rural areas, while maintaining strong spatial continuity. Furthermore, we found that urbanization is the primary factor exacerbating this inequality, while compact urban development can mitigate it. The findings of this study can provide practical guidance for cross-county ecological coordination, ecological restoration, and sustainable urban development.
- Open Access
- 16 September 2022
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 19 (4) , 1-26
An Application-oblivious Memory Scheduling System for DNN Accelerators
Deep Neural Networks (DNNs) tend to go deeper and wider, which poses a significant challenge to the training of DNNs, due to the limited memory capacity of DNN accelerators. Existing solutions for memory-efficient DNN training are densely coupled with the application features of DNN workloads, e.g., layer structures or computational graphs of DNNs are necessary for these solutions. This would result in weak versatility for DNNs with sophisticated layer structures or complicated computation graphs. These schemes usually need to be re-implemented or re-adapted due to the new layer structures or the unusual operators in the computational graphs introduced by these DNNs. In this article, we review the memory pressure issues of DNN training from the perspective of runtime systems and model the memory access behaviors of DNN workloads. We identify the iterative, regularity , and extremalization properties of memory access patterns for DNN workloads. Based on these observations, we propose AppObMem, an application-oblivious memory scheduling system. AppObMem automatically traces the memory behaviors of DNN workloads and schedules the memory swapping to reduce the memory pressure of the device accelerators without the perception of high-level information of layer structures or computation graphs. Evaluations on a variety of DNN models show that, AppObMem obtains 40–60% memory savings with acceptable performance loss. AppObMem is also competitive with other open sourced SOTA schemes.
- 16 September 2022
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 19 (4) , 1-26
Open AccessAn Application-oblivious Memory Scheduling System for DNN Accelerators
Deep Neural Networks (DNNs) tend to go deeper and wider, which poses a significant challenge to the training of DNNs, due to the limited memory capacity of DNN accelerators. Existing solutions for memory-efficient DNN training are densely coupled with the application features of DNN workloads, e.g., layer structures or computational graphs of DNNs are necessary for these solutions. This would result in weak versatility for DNNs with sophisticated layer structures or complicated computation graphs. These schemes usually need to be re-implemented or re-adapted due to the new layer structures or the unusual operators in the computational graphs introduced by these DNNs. In this article, we review the memory pressure issues of DNN training from the perspective of runtime systems and model the memory access behaviors of DNN workloads. We identify the iterative, regularity , and extremalization properties of memory access patterns for DNN workloads. Based on these observations, we propose AppObMem, an application-oblivious memory scheduling system. AppObMem automatically traces the memory behaviors of DNN workloads and schedules the memory swapping to reduce the memory pressure of the device accelerators without the perception of high-level information of layer structures or computation graphs. Evaluations on a variety of DNN models show that, AppObMem obtains 40–60% memory savings with acceptable performance loss. AppObMem is also competitive with other open sourced SOTA schemes.
- 25 March 2021
- Published by Springer Nature in International Journal of Parallel Programming
- Vol. 49 (5) , 628-645
Compiler-assisted Operator Template Library for DNN Accelerators
- 25 March 2021
- Published by Springer Nature in International Journal of Parallel Programming
- Vol. 49 (5) , 628-645
Compiler-assisted Operator Template Library for DNN Accelerators
- 18 January 2021
- Published by Springer Nature in Environment, Development and Sustainability
- Vol. 23 (8) , 11179-11200
Exploring the interactive coupled relationship between urban construction and resource environment in Wuhan, China
- 18 January 2021
- Published by Springer Nature in Environment, Development and Sustainability
- Vol. 23 (8) , 11179-11200
Exploring the interactive coupled relationship between urban construction and resource environment in Wuhan, China
- Open Access
Accelerating Deep Learning Inference with Cross-Layer Data Reuse on GPUs
Accelerating the deep learning inference is very important for real-time applications. In this paper, we propose a novel method to fuse the layers of convolutional neural networks (CNNs) on Graphics Processing Units (GPUs), which applies data reuse analysis and access optimization in different levels of the memory hierarchy. To achieve the balance between computation and memory access, we explore the fusion opportunities in the CNN computation graph and propose three fusion modes of convolutional neural networks: straight, merge and split. Then, an approach for generating efficient fused code is designed, which goes deeper in multi-level memory usage for cross-layer data reuse. The effectiveness of our method is evaluated with the network layers from state-of-the-art CNNs on two different GPU platforms, NVIDIA TITAN Xp and Tesla P4. The experiments show that the average speedup is 2.02x on representative structures of CNNs, and 1.57x on end-to-end inference of SqueezeNet.
Open AccessAccelerating Deep Learning Inference with Cross-Layer Data Reuse on GPUs
Accelerating the deep learning inference is very important for real-time applications. In this paper, we propose a novel method to fuse the layers of convolutional neural networks (CNNs) on Graphics Processing Units (GPUs), which applies data reuse analysis and access optimization in different levels of the memory hierarchy. To achieve the balance between computation and memory access, we explore the fusion opportunities in the CNN computation graph and propose three fusion modes of convolutional neural networks: straight, merge and split. Then, an approach for generating efficient fused code is designed, which goes deeper in multi-level memory usage for cross-layer data reuse. The effectiveness of our method is evaluated with the network layers from state-of-the-art CNNs on two different GPU platforms, NVIDIA TITAN Xp and Tesla P4. The experiments show that the average speedup is 2.02x on representative structures of CNNs, and 1.57x on end-to-end inference of SqueezeNet.
- 29 May 2020
- Published by Springer Nature in The Journal of Antibiotics
- Vol. 73 (9) , 642-645
Two novel milbemycin derivatives from the genetically engineered strain Streptomyces avermitilis AVE-H39
Two novel milbemycin derivatives, 5,27-epoxy-13α-hydroxy milbemycin β11 (1) and 5,27-epoxy-13α-hydroxy-25-ethyl milbemycin β11 (2), were isolated from the genetically engineered strain Streptomyces avermitilis AVE-H39. Their structures were elucidated through the interpretation of HR-ESIMS and extensive NMR spectroscopic data. Compounds 1 and 2 exhibited moderate acaricidal and nematicidal activities.
- 29 May 2020
- Published by Springer Nature in The Journal of Antibiotics
- Vol. 73 (9) , 642-645
Two novel milbemycin derivatives from the genetically engineered strain Streptomyces avermitilis AVE-H39
Two novel milbemycin derivatives, 5,27-epoxy-13α-hydroxy milbemycin β11 (1) and 5,27-epoxy-13α-hydroxy-25-ethyl milbemycin β11 (2), were isolated from the genetically engineered strain Streptomyces avermitilis AVE-H39. Their structures were elucidated through the interpretation of HR-ESIMS and extensive NMR spectroscopic data. Compounds 1 and 2 exhibited moderate acaricidal and nematicidal activities.
- Open Access
The Pitfall of Evaluating Performance on Emerging AI Accelerators
In recent years, domain-specific hardware has brought significant performance improvements in deep learning (DL). Both industry and academia only focus on throughput when evaluating these AI accelerators, which usually are custom ASICs deployed in datacenter to speed up the inference phase of DL workloads. Pursuing higher hardware throughput such as OPS (Operation Per Second) using various optimizations seems to be their main design target. However, they ignore the importance of accuracy in the DL nature. Motivated by this, this paper argue that a single throughput metric can not comprehensively reflect the real-world performance of AI accelerators. To reveal this pitfall, we evaluates several frequently-used optimizations on a typical AI accelerator and quantifies their impact on accuracy and throughout under representative DL inference workloads. Based on our experimental results, we find that some optimizations cause significant loss on accuracy in some workloads, although it can improves the throughout. Furthermore, our results show the importance of end-to-end evaluation in DL.
Open AccessThe Pitfall of Evaluating Performance on Emerging AI Accelerators
In recent years, domain-specific hardware has brought significant performance improvements in deep learning (DL). Both industry and academia only focus on throughput when evaluating these AI accelerators, which usually are custom ASICs deployed in datacenter to speed up the inference phase of DL workloads. Pursuing higher hardware throughput such as OPS (Operation Per Second) using various optimizations seems to be their main design target. However, they ignore the importance of accuracy in the DL nature. Motivated by this, this paper argue that a single throughput metric can not comprehensively reflect the real-world performance of AI accelerators. To reveal this pitfall, we evaluates several frequently-used optimizations on a typical AI accelerator and quantifies their impact on accuracy and throughout under representative DL inference workloads. Based on our experimental results, we find that some optimizations cause significant loss on accuracy in some workloads, although it can improves the throughout. Furthermore, our results show the importance of end-to-end evaluation in DL.
- Open Access
- 15 May 2019
- Published by MDPI AG in Sustainability
- Vol. 11 (10) , 2786
Exploring the Spatial Pattern and Influencing Factors of Land Carrying Capacity in Wuhan
Land carrying capacity is an important factor for urban sustainable development. It provides essential insights into land resource allocation and management. In this article, we propose a framework to evaluate land carrying capacity with multiple data sources from the first geographical census and socioeconomic statistics. In particular, an index, Land Resource Pressure (LRP), is proposed to evaluate the land carrying capacity, and a case study was carried out in Wuhan. The LRP of Wuhan was calculated on 250 m * 250 m grids, and showed a circularly declining pattern from central to outer areas. We collected its influencing factors in terms of nature resources, economy, transportation and urban construction, and then analyzed its causes via geographically weighted (GW) models. Firstly, pair-wise correlations between LRP and each influencing factor were explored via the GW correlation coefficients. These local estimates provide an important precursor for the following quantitative analysis via the GW regression (GWR) technique. The GWR coefficient estimates interpret the influences on LRP in a localized view. Results show that per capita gross domestic product (PerGDP) showed a higher absolute estimate among all factors, which proves that PerGDP has a relieving effect on LRP, especially in the southwestern areas. Overall, this study provides a technical framework to evaluate land carrying capacity with multi-source data sets and explore its localized influences via GW models, which could provide practical guidance for similar studies in other cities.
- 15 May 2019
- Published by MDPI AG in Sustainability
- Vol. 11 (10) , 2786
Open AccessExploring the Spatial Pattern and Influencing Factors of Land Carrying Capacity in Wuhan
Land carrying capacity is an important factor for urban sustainable development. It provides essential insights into land resource allocation and management. In this article, we propose a framework to evaluate land carrying capacity with multiple data sources from the first geographical census and socioeconomic statistics. In particular, an index, Land Resource Pressure (LRP), is proposed to evaluate the land carrying capacity, and a case study was carried out in Wuhan. The LRP of Wuhan was calculated on 250 m * 250 m grids, and showed a circularly declining pattern from central to outer areas. We collected its influencing factors in terms of nature resources, economy, transportation and urban construction, and then analyzed its causes via geographically weighted (GW) models. Firstly, pair-wise correlations between LRP and each influencing factor were explored via the GW correlation coefficients. These local estimates provide an important precursor for the following quantitative analysis via the GW regression (GWR) technique. The GWR coefficient estimates interpret the influences on LRP in a localized view. Results show that per capita gross domestic product (PerGDP) showed a higher absolute estimate among all factors, which proves that PerGDP has a relieving effect on LRP, especially in the southwestern areas. Overall, this study provides a technical framework to evaluate land carrying capacity with multi-source data sets and explore its localized influences via GW models, which could provide practical guidance for similar studies in other cities.
- 7 December 2018
- Published by Springer Nature in International Journal of Parallel Programming
- Vol. 47 (3) , 520-534
ElasticActor: An Actor System with Automatic Granularity Adjustment
- 7 December 2018
- Published by Springer Nature in International Journal of Parallel Programming
- Vol. 47 (3) , 520-534
ElasticActor: An Actor System with Automatic Granularity Adjustment