Ji Qi
Research Subjects
Research Interests
Ji Qi has a total of 56 co-authors. Here are the top co-authors by number of shared publications:
- Hongli Xu
School of Computer Science and Technology, University of Science and Technology of China, Hefei, Anhui, China
2 shared publications - Weichen Dai
University of Science and Technology of China
2 shared publications - Xiangyu Zhou
China Mobile (Suzhou) Software Technology Co., Ltd
2 shared publications - Yi Zhou
School of Information Science and Technology National Engineering Laboratory for Brain-inspired Intelligence Technology and Applications USTC
2 shared publications - Bao Sun
China Mobile (Suzhou) Software Technology Co., Ltd, China
1 shared publication - Baoqing Wang
School of Computer Science and Technology, University of Science and Technology of China, Hefei, Anhui, China
1 shared publication
Latest Publications
- 4 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. PP (00189340) , 1-17
Scalable High-fidelity Cloud Network Validation via Hybrid Architecture
Ensuring reliable operation of cloud networks is critical for cloud service providers to guarantee quality of service for tenants. A promising solution is to design a high-fidelity cloud network validation platform that proactively validates the correctness of all operations before implementing changes to the production network. However, the tight coupling between physical and virtual networks in the cloud poses challenges to achieving high-fidelity cloud network validation. Existing network validation platforms focus primarily on traditional physical networks, while ignoring virtual network validation. Regrettably, neglecting the combined validation of physical and virtual networks will result in inaccurate evaluations. To bridge this gap, we present HifiCNet, a high-fidelity platform that concurrently validates both physical and virtual networks. HifiCNet designs an orchestrator to elegantly coordinate the interaction between physical and virtual networks in the cloud and innovatively adopts an emulator-simulator hybrid architecture to ensure high fidelity and scalability for cloud network validation. Through extensive evaluation based on real topologies and traffic traces, we show that HifiCNet enables high-fidelity validation of cloud network configurations, services, and exceptions. Notably, HifiCNet can leverage 38 servers to establish a physical network comprising 10k hosts, as well as a virtual network consisting of 200k virtual machines.
- 4 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. PP (00189340) , 1-17
Scalable High-fidelity Cloud Network Validation via Hybrid Architecture
Ensuring reliable operation of cloud networks is critical for cloud service providers to guarantee quality of service for tenants. A promising solution is to design a high-fidelity cloud network validation platform that proactively validates the correctness of all operations before implementing changes to the production network. However, the tight coupling between physical and virtual networks in the cloud poses challenges to achieving high-fidelity cloud network validation. Existing network validation platforms focus primarily on traditional physical networks, while ignoring virtual network validation. Regrettably, neglecting the combined validation of physical and virtual networks will result in inaccurate evaluations. To bridge this gap, we present HifiCNet, a high-fidelity platform that concurrently validates both physical and virtual networks. HifiCNet designs an orchestrator to elegantly coordinate the interaction between physical and virtual networks in the cloud and innovatively adopts an emulator-simulator hybrid architecture to ensure high fidelity and scalability for cloud network validation. Through extensive evaluation based on real topologies and traffic traces, we show that HifiCNet enables high-fidelity validation of cloud network configurations, services, and exceptions. Notably, HifiCNet can leverage 38 servers to establish a physical network comprising 10k hosts, as well as a virtual network consisting of 200k virtual machines.
- 9 November 2025
- Published by Springer Nature
- p. 19-33
DGFN: Brain-Inspired Distributed Fusion for Robust Multimodal Sentiment Analysis
- 9 November 2025
- Published by Springer Nature
- p. 19-33
DGFN: Brain-Inspired Distributed Fusion for Robust Multimodal Sentiment Analysis
- Open Access
- 6 November 2025
- Published by MDPI AG in Electronics
- Vol. 14 (21) , 4345
Renewable-Aware Container Migration in Multi-Data Centers
The proliferation of artificial intelligence (AI) and online services has significantly escalated the demand for computing and storage resources, which are fundamentally enabled by cloud computing infrastructure. As the backbone of cloud computing services, data centers have undergone continuous expansion in scale, consequently leading to significant energy consumption and a significant carbon footprint. To effectively mitigate the environmental impact, the strategy should prioritize the integration of renewable energy, while simultaneously minimizing other contributing factors such as energy consumption. Achieving both objectives simultaneously requires a fine-grained, dynamic approach to workload management. To this end, this study proposes a comprehensive container placement strategy that integrates a dynamic priority-based container selection algorithm with a multi-factor single-objective container placement algorithm based on the Dream Optimization Algorithm (DOA). The placement algorithm converts multiple factors—including load balancing in multi-data center environments, energy consumption, renewable energy utilization rate, carbon emissions, Service Level Agreement Violation (SLAV), and container migration costs—into a comprehensive fitness metric. Experimental results on Google and Alibaba datasets show our method consistently achieves the highest renewable energy utilization rates (up to 92.08%) and the lowest carbon emissions. Furthermore, our integrated strategy demonstrates a superior trade-off, reducing migration counts by up to 16.3% and SLAV by 12.4% compared to baselines, while maintaining excellent green performance. This establishes our method as a practical and effective solution for sustainable cloud computing.
- 6 November 2025
- Published by MDPI AG in Electronics
- Vol. 14 (21) , 4345
Open AccessRenewable-Aware Container Migration in Multi-Data Centers
The proliferation of artificial intelligence (AI) and online services has significantly escalated the demand for computing and storage resources, which are fundamentally enabled by cloud computing infrastructure. As the backbone of cloud computing services, data centers have undergone continuous expansion in scale, consequently leading to significant energy consumption and a significant carbon footprint. To effectively mitigate the environmental impact, the strategy should prioritize the integration of renewable energy, while simultaneously minimizing other contributing factors such as energy consumption. Achieving both objectives simultaneously requires a fine-grained, dynamic approach to workload management. To this end, this study proposes a comprehensive container placement strategy that integrates a dynamic priority-based container selection algorithm with a multi-factor single-objective container placement algorithm based on the Dream Optimization Algorithm (DOA). The placement algorithm converts multiple factors—including load balancing in multi-data center environments, energy consumption, renewable energy utilization rate, carbon emissions, Service Level Agreement Violation (SLAV), and container migration costs—into a comprehensive fitness metric. Experimental results on Google and Alibaba datasets show our method consistently achieves the highest renewable energy utilization rates (up to 92.08%) and the lowest carbon emissions. Furthermore, our integrated strategy demonstrates a superior trade-off, reducing migration counts by up to 16.3% and SLAV by 12.4% compared to baselines, while maintaining excellent green performance. This establishes our method as a practical and effective solution for sustainable cloud computing.
- Open Access
DiLoCoX: A Low-Communication Large-Scale Training Framework for Decentralized Cluster
The distributed training of foundation models, particularly large language models (LLMs), demands a high level of communication. Consequently, it is highly dependent on a centralized cluster with fast and reliable interconnects. Can we conduct training on slow networks and thereby unleash the power of decentralized clusters when dealing with models exceeding 100 billion parameters? In this paper, we propose DiLoCoX, a low-communication large-scale decentralized cluster training framework. It combines Pipeline Parallelism with Dual Optimizer Policy, One-Step-Delay Overlap of Communication and Local Training, and an Adaptive Gradient Compression Scheme. This combination significantly improves the scale of parameters and the speed of model pre-training. We justify the benefits of one-step-delay overlap of communication and local training, as well as the adaptive gradient compression scheme, through a theoretical analysis of convergence. Empirically, we demonstrate that DiLoCoX is capable of pre-training a 107B foundation model over a 1Gbps network. Compared to vanilla AllReduce, DiLoCoX can achieve a 357x speedup in distributed training while maintaining negligible degradation in model convergence. To the best of our knowledge, this is the first decentralized training framework successfully applied to models with over 100 billion parameters.
Open AccessDiLoCoX: A Low-Communication Large-Scale Training Framework for Decentralized Cluster
The distributed training of foundation models, particularly large language models (LLMs), demands a high level of communication. Consequently, it is highly dependent on a centralized cluster with fast and reliable interconnects. Can we conduct training on slow networks and thereby unleash the power of decentralized clusters when dealing with models exceeding 100 billion parameters? In this paper, we propose DiLoCoX, a low-communication large-scale decentralized cluster training framework. It combines Pipeline Parallelism with Dual Optimizer Policy, One-Step-Delay Overlap of Communication and Local Training, and an Adaptive Gradient Compression Scheme. This combination significantly improves the scale of parameters and the speed of model pre-training. We justify the benefits of one-step-delay overlap of communication and local training, as well as the adaptive gradient compression scheme, through a theoretical analysis of convergence. Empirically, we demonstrate that DiLoCoX is capable of pre-training a 107B foundation model over a 1Gbps network. Compared to vanilla AllReduce, DiLoCoX can achieve a 357x speedup in distributed training while maintaining negligible degradation in model convergence. To the best of our knowledge, this is the first decentralized training framework successfully applied to models with over 100 billion parameters.
- 26 June 2025
- Published by Springer Nature
- p. 455-469
Exploring the Neural Burden in Pruned Models: An Insight Inspired by Neuroscience
- 26 June 2025
- Published by Springer Nature
- p. 455-469
Exploring the Neural Burden in Pruned Models: An Insight Inspired by Neuroscience
- 10 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 33 (4) , 1438-1453
Adaptive Local Update and Neural Composition for Accelerating Federated Learning in Heterogeneous Edge Networks
Federated Learning (FL) enables distributed clients to collaboratively train models without exposing their private data. However, it is difficult to implement efficient FL due to limited resources. Most existing works compress the transmitted gradients or prune the global model to reduce the resource cost, but leave the compressed or pruned parameters under-optimized, which degrades the training performance. To address this issue, the neural composition technique constructs size-adjustable models by composing low-rank tensors, allowing every parameter in the global model to learn the knowledge from all clients. Nevertheless, some tensors can only be optimized by a small fraction of clients, thus the global model may get insufficient training, leading to a long completion time, especially in heterogeneous edge scenarios. To this end, we enhance the neural composition technique, enabling all parameters to be fully trained. Further, we propose a lightweight FL framework, called Heroes, with enhanced neural composition and adaptive local update. A greedy-based algorithm is designed to adaptively assign the proper tensors and local update frequencies for participating clients according to their heterogeneous capabilities and resource budgets. On this basis, we further propose an extension of Heroes, termed AdaHeroes, which further improves the training performance under the statistical heterogeneity scenario based on an adaptive client selection strategy. Extensive experiments demonstrate that Heroes can reduce traffic consumption by about 72.46% and provide up to 2.76× speedup compared to the baselines. Furthermore, with the setting of statistical heterogeneity, AdaHeroes can improve the test accuracy by about 4.77% compared with Heroes and the baselines.
- 10 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 33 (4) , 1438-1453
Adaptive Local Update and Neural Composition for Accelerating Federated Learning in Heterogeneous Edge Networks
Federated Learning (FL) enables distributed clients to collaboratively train models without exposing their private data. However, it is difficult to implement efficient FL due to limited resources. Most existing works compress the transmitted gradients or prune the global model to reduce the resource cost, but leave the compressed or pruned parameters under-optimized, which degrades the training performance. To address this issue, the neural composition technique constructs size-adjustable models by composing low-rank tensors, allowing every parameter in the global model to learn the knowledge from all clients. Nevertheless, some tensors can only be optimized by a small fraction of clients, thus the global model may get insufficient training, leading to a long completion time, especially in heterogeneous edge scenarios. To this end, we enhance the neural composition technique, enabling all parameters to be fully trained. Further, we propose a lightweight FL framework, called Heroes, with enhanced neural composition and adaptive local update. A greedy-based algorithm is designed to adaptively assign the proper tensors and local update frequencies for participating clients according to their heterogeneous capabilities and resource budgets. On this basis, we further propose an extension of Heroes, termed AdaHeroes, which further improves the training performance under the statistical heterogeneity scenario based on an adaptive client selection strategy. Extensive experiments demonstrate that Heroes can reduce traffic consumption by about 72.46% and provide up to 2.76× speedup compared to the baselines. Furthermore, with the setting of statistical heterogeneity, AdaHeroes can improve the test accuracy by about 4.77% compared with Heroes and the baselines.
- 17 February 2025
- Published by Springer Nature
- p. 179-199
Low-Carbon Geographically Distributed Cloud-Edge Task Scheduling
- 17 February 2025
- Published by Springer Nature
- p. 179-199
Low-Carbon Geographically Distributed Cloud-Edge Task Scheduling
- 11 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 17 (5) , 2248-2260
Diner: Interpretable Anomaly Detection for Seasonal Time Series in Web Services
Monitoring and anomaly detection of key performance indicators (KPIs) are crucial for large Internet companies to maintain the reliability of their Web services. Influenced by human behavior and schedules, the KPIs of Web services typically exhibit seasonal characteristics. These characteristics may be complex as different KPIs exhibit differences in trend, multiple periods, and noise behaviors. However, existing anomaly detection methods typically only model one fixed pattern of seasonal KPIs, which may lead to performance degradation when dealing with diverse seasonal KPIs. In this work, we propose a novel anomaly detection model for seasonal KPIs, Diner , which incorporates multiple interpretable components. It is able to capture the additive and multiplicative trends, multiple periods, and seasonal noise in intricate seasonal KPIs, making it easily adaptable to different types of seasonal KPIs. Additionally, we present a set of evaluation criteria for generic time series anomaly detection tasks, which prove more effective in handling ambiguous manual labels and various anomaly events. Experiments are conducted on three real-world datasets, and the performance Diner surpassed both the statistical baseline and the state-of-the-art deep learning baselines.
- 11 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 17 (5) , 2248-2260
Diner: Interpretable Anomaly Detection for Seasonal Time Series in Web Services
Monitoring and anomaly detection of key performance indicators (KPIs) are crucial for large Internet companies to maintain the reliability of their Web services. Influenced by human behavior and schedules, the KPIs of Web services typically exhibit seasonal characteristics. These characteristics may be complex as different KPIs exhibit differences in trend, multiple periods, and noise behaviors. However, existing anomaly detection methods typically only model one fixed pattern of seasonal KPIs, which may lead to performance degradation when dealing with diverse seasonal KPIs. In this work, we propose a novel anomaly detection model for seasonal KPIs, Diner , which incorporates multiple interpretable components. It is able to capture the additive and multiplicative trends, multiple periods, and seasonal noise in intricate seasonal KPIs, making it easily adaptable to different types of seasonal KPIs. Additionally, we present a set of evaluation criteria for generic time series anomaly detection tasks, which prove more effective in handling ambiguous manual labels and various anomaly events. Experiments are conducted on three real-world datasets, and the performance Diner surpassed both the statistical baseline and the state-of-the-art deep learning baselines.
- Open Access
- 27 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 49-55
Virtualized Network Function Scaling Strategy Based on Additional Momentum BP Neural Network
With the increasing demand for data, mobile operator face the challenge of resource reuse and cost reduction, while also needing to improve network performance and support dynamic scaling. In this paper, we propose a Virtualized Network Function (VNF) scaling ratio prediction model based on an additional momentum BP neural network to optimize the performance of VNF instances in the context of public cloud Network Function Virtualization (NFV) scenarios. We model performance monitoring data and optimize the BP neural network model using the additional momentum method, resulting in specific VNF scaling strategies. The results show that the additional momentum BP neural network is highly effective in predicting the number of VNF scaling instances, with an error rate of only 0.0009 and an accuracy of 98%. Compared to the unimproved model, the additional momentum method leads to smaller errors and notable improvements in both convergence speed and accuracy.
- 27 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 49-55
Open AccessVirtualized Network Function Scaling Strategy Based on Additional Momentum BP Neural Network
With the increasing demand for data, mobile operator face the challenge of resource reuse and cost reduction, while also needing to improve network performance and support dynamic scaling. In this paper, we propose a Virtualized Network Function (VNF) scaling ratio prediction model based on an additional momentum BP neural network to optimize the performance of VNF instances in the context of public cloud Network Function Virtualization (NFV) scenarios. We model performance monitoring data and optimize the BP neural network model using the additional momentum method, resulting in specific VNF scaling strategies. The results show that the additional momentum BP neural network is highly effective in predicting the number of VNF scaling instances, with an error rate of only 0.0009 and an accuracy of 98%. Compared to the unimproved model, the additional momentum method leads to smaller errors and notable improvements in both convergence speed and accuracy.
- Open Access
- 15 September 2021
- Published by Wiley in Journal of Advanced Transportation
- Vol. 2021, 1-16
A Novel Convex Hull Coverage Algorithm for the Deployment of Electric Taxi Swap Stations Based on Urban Traffic Flow
Electric taxis have been adopted as a new energy public transportation tool as opposed to traditional taxis in modern city. Designing an efficient swapping station deployment scheme has become an important issue to improve the endurance capability of electric taxis. In this study, based on real operation trajectory data from 3997 taxis in Suzhou city, the battery swapping demand of taxis based on urban traffic flow is obtained to construct a network coverage deployment model under rule constraints, where the main optimization goal is to minimize the number of swapping stations and load balancing. According to this model, a traffic drive planning algorithm based on computational geometry is presented. The experimental results illustrate that the deployment scheme obtained by the proposed algorithm is significantly optimized in terms of the deployment cost and service load and has a lower algorithm time complexity than the typical unified deployment scheme. Therefore, the proposed method can be applied to improve the operating efficiency of the urban electric taxi system.
- 15 September 2021
- Published by Wiley in Journal of Advanced Transportation
- Vol. 2021, 1-16
Open AccessA Novel Convex Hull Coverage Algorithm for the Deployment of Electric Taxi Swap Stations Based on Urban Traffic Flow
Electric taxis have been adopted as a new energy public transportation tool as opposed to traditional taxis in modern city. Designing an efficient swapping station deployment scheme has become an important issue to improve the endurance capability of electric taxis. In this study, based on real operation trajectory data from 3997 taxis in Suzhou city, the battery swapping demand of taxis based on urban traffic flow is obtained to construct a network coverage deployment model under rule constraints, where the main optimization goal is to minimize the number of swapping stations and load balancing. According to this model, a traffic drive planning algorithm based on computational geometry is presented. The experimental results illustrate that the deployment scheme obtained by the proposed algorithm is significantly optimized in terms of the deployment cost and service load and has a lower algorithm time complexity than the typical unified deployment scheme. Therefore, the proposed method can be applied to improve the operating efficiency of the urban electric taxi system.