Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010090
Kaiyuan Yang;Longchao Liu;Haotian Liu;Tiantai Deng
Transforming complex problems, such as transforming ordinary differential equations (ODEs) into matrix formats, into simpler computational tasks is key for AI advancements and paves the way for more efficient computing architectures. Systolic Arrays, known for their computational efficiency, low power use and ease of implementation, address AI's computational challenges. They are central to mainstream industry AI accelerators, with improvements to the Processing Element (PE) significantly boosting systolic array performance, and also streamlines computing architectures, paving the way for more efficient solutions in technology fields. This research presents a novel PE design and its integration of systolic array based on a novel computing theory - bit-level mathematics for Multiply-Accumulate (MAC) operation. We present 3 different architectures for the PE and provide a comprehensive comparison between them and the state-of-the-art technologies, focusing on power, area, and throughput. This research also demonstrates the integration of the proposed MAC unit design with systolic arrays, highlighting significant improvements in computational efficiency. Our implementations show a 2380952.38 times lower latency, yet 64.19 times less DSP48E1, 1.26 times less Look-Up Tables (LUTs), 10.76 times less Flip-Flops (FFs), with 99.63 times less power consumption and 15.19 times higher performance per PE compared to the state-of-the-art design.
{"title":"A Novel Parallel Processing Element Architecture for Accelerating ODE and AI","authors":"Kaiyuan Yang;Longchao Liu;Haotian Liu;Tiantai Deng","doi":"10.26599/TST.2024.9010090","DOIUrl":"https://doi.org/10.26599/TST.2024.9010090","url":null,"abstract":"Transforming complex problems, such as transforming ordinary differential equations (ODEs) into matrix formats, into simpler computational tasks is key for AI advancements and paves the way for more efficient computing architectures. Systolic Arrays, known for their computational efficiency, low power use and ease of implementation, address AI's computational challenges. They are central to mainstream industry AI accelerators, with improvements to the Processing Element (PE) significantly boosting systolic array performance, and also streamlines computing architectures, paving the way for more efficient solutions in technology fields. This research presents a novel PE design and its integration of systolic array based on a novel computing theory - bit-level mathematics for Multiply-Accumulate (MAC) operation. We present 3 different architectures for the PE and provide a comprehensive comparison between them and the state-of-the-art technologies, focusing on power, area, and throughput. This research also demonstrates the integration of the proposed MAC unit design with systolic arrays, highlighting significant improvements in computational efficiency. Our implementations show a 2380952.38 times lower latency, yet 64.19 times less DSP48E1, 1.26 times less Look-Up Tables (LUTs), 10.76 times less Flip-Flops (FFs), with 99.63 times less power consumption and 15.19 times higher performance per PE compared to the state-of-the-art design.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"1954-1964"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979797","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888407","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010184
Rui Li;Hai Wang;Qiang Lu;Jie Yan;Shuo Ji;Yuhui Ma
Federated learning (FL) technology has significant advantages in solving data silos and user privacy problems, but the traditional federal average (FedAvg) algorithm is ineffective in classifying and faces the risk of refactoring attacks when dealing with non-independent and identically distributed (Non-IID) data, which is especially prominent since medical data involves sensitive personal health information. Therefore, optimizing FedAvg to adapt to Non-IID data distribution and enhancing privacy protection are urgent problems that need to be solved, and this paper investigates these two aspects. In order to enhance the classification performance of FedAvg under Non-IID distribution, this paper combines the optimized deep learning model SE-ResNet18-E with FedAvg to obtain the FedAvg(SE-ResNet18-E) algorithm. The algorithm takes advantage of the SE-ResNet18-E model in feature extraction and classification tasks, fully uses the data resources of each participant, and improves the classification performance of FedAvg under Non-IID distribution. In addition, the algorithm achieves high communication performance. Second, in order to enhance the security of FL in the medical domain, threshold Paillier encryption is further introduced on top of FedAvg(SE-ResNet18-E) to form the Safe-FedAvg(SE-ResNet18-E) algorithm, which solves the threat of reconstruction attack and private key leakage in medical FL. After experimental validation, the Safe-FedAvg (SE-ResNet18-E) algorithm effectively improves the accuracy of disease classification and effectively protects the privacy and security of medical data, and enhances the trust of medical organizations participating in FL.
{"title":"Research on Medical Image Classification Based on Improved FedAvg Algorithm","authors":"Rui Li;Hai Wang;Qiang Lu;Jie Yan;Shuo Ji;Yuhui Ma","doi":"10.26599/TST.2024.9010184","DOIUrl":"https://doi.org/10.26599/TST.2024.9010184","url":null,"abstract":"Federated learning (FL) technology has significant advantages in solving data silos and user privacy problems, but the traditional federal average (FedAvg) algorithm is ineffective in classifying and faces the risk of refactoring attacks when dealing with non-independent and identically distributed (Non-IID) data, which is especially prominent since medical data involves sensitive personal health information. Therefore, optimizing FedAvg to adapt to Non-IID data distribution and enhancing privacy protection are urgent problems that need to be solved, and this paper investigates these two aspects. In order to enhance the classification performance of FedAvg under Non-IID distribution, this paper combines the optimized deep learning model SE-ResNet18-E with FedAvg to obtain the FedAvg(SE-ResNet18-E) algorithm. The algorithm takes advantage of the SE-ResNet18-E model in feature extraction and classification tasks, fully uses the data resources of each participant, and improves the classification performance of FedAvg under Non-IID distribution. In addition, the algorithm achieves high communication performance. Second, in order to enhance the security of FL in the medical domain, threshold Paillier encryption is further introduced on top of FedAvg(SE-ResNet18-E) to form the Safe-FedAvg(SE-ResNet18-E) algorithm, which solves the threat of reconstruction attack and private key leakage in medical FL. After experimental validation, the Safe-FedAvg (SE-ResNet18-E) algorithm effectively improves the accuracy of disease classification and effectively protects the privacy and security of medical data, and enhances the trust of medical organizations participating in FL.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2243-2258"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979789","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010154
Anjie Peng;Guoqiang Shi;Zhi Lin;Hui Zeng;Xing Yang
Adversarial attacks craft adversarial examples (AEs) to fool convolution neural networks. The mainstream gradient-based attacks, based on first-order optimization methods, encounter bottlenecks to generate high transferable AEs attacking unknown models. Considering that the high-order method would be a better optimization algorithm, we attempt to build high-order adversarial attacks to improve the transferability of AEs. However, solving the optimization problem of adversarial attacks directly via higher-order derivatives is computationally difficult and may face the non-convergence problem. So, we leverage the Runge-Kutta (RK) method, which is an accurate yet efficient high-order numerical solver of ordinary differential equation (ODE), to approximate high-order adversarial attacks. We first induce the gradient descent process of gradient-based attack as an ODE, and then numerically solve the ODE via RK method to develop approximated high-order adversarial attacks. Concretely, through ignoring the higher-order infinitesimal item in the Taylor expansion of the loss, the proposed method utilizes a linear combination of the present gradient and looking-ahead gradients to replace the computationally expensive high-order derivatives, and yields a relatively fast equivalent high-order adversarial attack. The proposed high-order adversarial attack can be extensively integrated with transferability augmentation methods to generate high transferable AEs. Extensive experiments demonstrate that the RK-based attacks exhibit higher transferability than the state of the arts.
{"title":"Approximating High-Order Adversarial Attacks Using Runge-Kutta Methods","authors":"Anjie Peng;Guoqiang Shi;Zhi Lin;Hui Zeng;Xing Yang","doi":"10.26599/TST.2024.9010154","DOIUrl":"https://doi.org/10.26599/TST.2024.9010154","url":null,"abstract":"Adversarial attacks craft adversarial examples (AEs) to fool convolution neural networks. The mainstream gradient-based attacks, based on first-order optimization methods, encounter bottlenecks to generate high transferable AEs attacking unknown models. Considering that the high-order method would be a better optimization algorithm, we attempt to build high-order adversarial attacks to improve the transferability of AEs. However, solving the optimization problem of adversarial attacks directly via higher-order derivatives is computationally difficult and may face the non-convergence problem. So, we leverage the Runge-Kutta (RK) method, which is an accurate yet efficient high-order numerical solver of ordinary differential equation (ODE), to approximate high-order adversarial attacks. We first induce the gradient descent process of gradient-based attack as an ODE, and then numerically solve the ODE via RK method to develop approximated high-order adversarial attacks. Concretely, through ignoring the higher-order infinitesimal item in the Taylor expansion of the loss, the proposed method utilizes a linear combination of the present gradient and looking-ahead gradients to replace the computationally expensive high-order derivatives, and yields a relatively fast equivalent high-order adversarial attack. The proposed high-order adversarial attack can be extensively integrated with transferability augmentation methods to generate high transferable AEs. Extensive experiments demonstrate that the RK-based attacks exhibit higher transferability than the state of the arts.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"1927-1939"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979816","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888330","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The YOLOv5 algorithm is widely used in edge computing systems for object detection. However, the limited computing resources of embedded devices and the large model size of existing deep learning based methods increase the difficulty of real-time object detection on edge devices. To address this issue, we propose a smaller, less computationally intensive, and more accurate algorithm for object detection. Multi-scale Feature Fusion-YOLO (MFF-YOLO) is built on top of the YOLOv5s framework, but it contains substantial improvements to YOLOv5s. First, we design the MFF module to improve the feature propagation path in the feature pyramid, which further integrates the semantic information from different paths of feature layers. Then, a large convolution-kernel module is used in the bottleneck. The structure enlarges the receptive field and preserves shallow semantic information, which overcomes the performance limitation arising from uneven propagation in Feature Pyramid Networks (FPN). In addition, a multi-branch downsampling method based on depthwise separable convolutions and a bottleneck structure with deformable convolutions are designed to reduce the complexity of the backbone network and minimize the real-time performance loss caused by the increased model complexity. The experimental results on PASCAL VOC and MS COCO datasets show that, compared with YOLOv5s, MFF-YOLO reduces the number of parameters by 7% and the number of FLoating point Operations Per second (FLOPs) by 11.8%. The mAP@0.5 has improved by 3.7% and 5.5%, and the mAP@0.5:0.95 has improved by 6.5% and 6.2%, respetively. Furthermore, compared with YOLOv7-tiny, PP-YOLO-tiny, and other mainstream methods, MFF-YOLO has achieved better results on multiple indicators.
{"title":"MFF-YOLO: An Improved YOLO Algorithm Based on Multi-Scale Semantic Feature Fusion","authors":"Junsan Zhang;Chenyang Xu;Shigen Shen;Jie Zhu;Peiying Zhang","doi":"10.26599/TST.2024.9010097","DOIUrl":"https://doi.org/10.26599/TST.2024.9010097","url":null,"abstract":"The YOLOv5 algorithm is widely used in edge computing systems for object detection. However, the limited computing resources of embedded devices and the large model size of existing deep learning based methods increase the difficulty of real-time object detection on edge devices. To address this issue, we propose a smaller, less computationally intensive, and more accurate algorithm for object detection. Multi-scale Feature Fusion-YOLO (MFF-YOLO) is built on top of the YOLOv5s framework, but it contains substantial improvements to YOLOv5s. First, we design the MFF module to improve the feature propagation path in the feature pyramid, which further integrates the semantic information from different paths of feature layers. Then, a large convolution-kernel module is used in the bottleneck. The structure enlarges the receptive field and preserves shallow semantic information, which overcomes the performance limitation arising from uneven propagation in Feature Pyramid Networks (FPN). In addition, a multi-branch downsampling method based on depthwise separable convolutions and a bottleneck structure with deformable convolutions are designed to reduce the complexity of the backbone network and minimize the real-time performance loss caused by the increased model complexity. The experimental results on PASCAL VOC and MS COCO datasets show that, compared with YOLOv5s, MFF-YOLO reduces the number of parameters by 7% and the number of FLoating point Operations Per second (FLOPs) by 11.8%. The mAP@0.5 has improved by 3.7% and 5.5%, and the mAP@0.5:0.95 has improved by 6.5% and 6.2%, respetively. Furthermore, compared with YOLOv7-tiny, PP-YOLO-tiny, and other mainstream methods, MFF-YOLO has achieved better results on multiple indicators.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2097-2113"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979796","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010105
Jiajie Xu;Haolong Xiang;Shaobo Zang;Muhammad Bilal;Maqbool Khan;Guangming Cui
Smart city pollution control is fundamental to urban sustainability, which relies extensively on physical infrastructure such as sensors and cameras for real-time monitoring. Generally, monitoring data needs to be transmitted to centralized servers for pollution control service determination. In order to achieve highly efficient service quality, edge computing is involved in the smart city pollution control system (SCPCS) as it provides computational capabilities near the monitoring devices and low-latency pollution control services. However, considering the diversity of service requests, determination of offloading destination is a crucial challenge for SCPCS. In this paper, A Deep Q-Network (DQN)-based edge offloading method, called N-DEO, is proposed. Initially, N-DEO employs neural hierarchical interpolation for time series forecasting (N-HITS) to forecast pollution control service requests. Afterwards, an epsilon-greedy policy is designed to select actions. Finally, the optimal service offloading strategy is determined by the DQN algorithm. Experimental results demonstrate that N-DEO achieves the higher performance on service latency and system load compared with the current state-of-the-art methods.
{"title":"A DQN-Based Edge Offloading Method for Smart City Pollution Control","authors":"Jiajie Xu;Haolong Xiang;Shaobo Zang;Muhammad Bilal;Maqbool Khan;Guangming Cui","doi":"10.26599/TST.2024.9010105","DOIUrl":"https://doi.org/10.26599/TST.2024.9010105","url":null,"abstract":"Smart city pollution control is fundamental to urban sustainability, which relies extensively on physical infrastructure such as sensors and cameras for real-time monitoring. Generally, monitoring data needs to be transmitted to centralized servers for pollution control service determination. In order to achieve highly efficient service quality, edge computing is involved in the smart city pollution control system (SCPCS) as it provides computational capabilities near the monitoring devices and low-latency pollution control services. However, considering the diversity of service requests, determination of offloading destination is a crucial challenge for SCPCS. In this paper, A Deep Q-Network (DQN)-based edge offloading method, called N-DEO, is proposed. Initially, N-DEO employs neural hierarchical interpolation for time series forecasting (N-HITS) to forecast pollution control service requests. Afterwards, an epsilon-greedy policy is designed to select actions. Finally, the optimal service offloading strategy is determined by the DQN algorithm. Experimental results demonstrate that N-DEO achieves the higher performance on service latency and system load compared with the current state-of-the-art methods.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2227-2242"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979785","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010085
Dayong Deng;Wenxin Shen;Zhixuan Deng;Tianrui Li;Anjin Liu
The ensemble learning model can effectively detect drift and utilize diversity to improve the performance of adapting to drift. However, local concept drift can occur in different types at different time points, causing basic learners are difficult to distinguish the drift of local boundaries, and the drift range is difficult to determine. Thus, the ensemble learning model to adapt local concept drifts is still challenging problem. Moreover, there are often differences in decision boundaries after drift adaptation, and employing overall diversity measurement is inappropriate. To address these two issues, this paper proposes a novel ensemble learning model called instance-weighted ensemble learning based on the three-way decision (IWE-TWD). In IWE-TWD, a divide-and-conquer strategy is employed to handle uncertain drift and to select base learners; Density clustering dynamically constructs density regions to lock drift range; Three-way decision is adopted to estimate whether the region distribution changes, and the instance is weighted with the probability of region distribution change; The diversities between base learners are determined with three-way decision also. Experimental results show that IWE-TWD has better performance than the state-of-the-art models in data stream classification on ten synthetic data sets and seven real-world data sets.
{"title":"An Ensemble Learning Model Based on Three-Way Decision for Concept Drift Adaptation","authors":"Dayong Deng;Wenxin Shen;Zhixuan Deng;Tianrui Li;Anjin Liu","doi":"10.26599/TST.2024.9010085","DOIUrl":"https://doi.org/10.26599/TST.2024.9010085","url":null,"abstract":"The ensemble learning model can effectively detect drift and utilize diversity to improve the performance of adapting to drift. However, local concept drift can occur in different types at different time points, causing basic learners are difficult to distinguish the drift of local boundaries, and the drift range is difficult to determine. Thus, the ensemble learning model to adapt local concept drifts is still challenging problem. Moreover, there are often differences in decision boundaries after drift adaptation, and employing overall diversity measurement is inappropriate. To address these two issues, this paper proposes a novel ensemble learning model called instance-weighted ensemble learning based on the three-way decision (IWE-TWD). In IWE-TWD, a divide-and-conquer strategy is employed to handle uncertain drift and to select base learners; Density clustering dynamically constructs density regions to lock drift range; Three-way decision is adopted to estimate whether the region distribution changes, and the instance is weighted with the probability of region distribution change; The diversities between base learners are determined with three-way decision also. Experimental results show that IWE-TWD has better performance than the state-of-the-art models in data stream classification on ten synthetic data sets and seven real-world data sets.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2029-2047"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979650","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In edge-distributed environments, spatiotemporal graphs provide a promising solution for capturing the complex dependencies among nodes and edges necessary for accurate wind speed forecasting. These dependencies involve spatial and temporal interactions that are crucial for modeling dynamic weather patterns. However, challenges, such as effectively maintaining spatial dependency information across spatiotemporal subgraphs, can lead to reduced prediction accuracy. Additionally, managing high communication costs, associated with the need for frequent and intensive data exchanges required for real-time forecasting across distributed nodes, poses significant hurdles. To address these issues, we propose graph coarsening-based cross-subgraph message passing with edge collaboration training mechanism (namely ComPact), a novel approach that simplifies graph structures through graph coarsening while preserving essential spatiotemporal dependencies. This coarsening process minimizes communication overhead and enables effective cross-subgraph message passing, capturing both local and long-range dependencies. ComPact further leverages hierarchical graph learning and structured edge collaboration to integrate global information into local subgraphs, enhancing predictive performance. Experimental validation on large-scale datasets, primarily the WindPower dataset, demonstrates ComPact's superiority in wind speed forecasting, with up to a 31.82% reduction in Mean Absolute Error (MAE) and 11.8% lower in Mean Absolute Percentage Error (MAPE) compared to federated learning baselines.
{"title":"ComPact: Edge Collaborative Spatiotemporal Graph Learning for Wind Speed Forecasting","authors":"Zaigang Gong;Siyu Chen;Qiangsheng Dai;Ying Feng;Jinghui Zhang","doi":"10.26599/TST.2024.9010261","DOIUrl":"https://doi.org/10.26599/TST.2024.9010261","url":null,"abstract":"In edge-distributed environments, spatiotemporal graphs provide a promising solution for capturing the complex dependencies among nodes and edges necessary for accurate wind speed forecasting. These dependencies involve spatial and temporal interactions that are crucial for modeling dynamic weather patterns. However, challenges, such as effectively maintaining spatial dependency information across spatiotemporal subgraphs, can lead to reduced prediction accuracy. Additionally, managing high communication costs, associated with the need for frequent and intensive data exchanges required for real-time forecasting across distributed nodes, poses significant hurdles. To address these issues, we propose graph coarsening-based cross-subgraph message passing with edge collaboration training mechanism (namely ComPact), a novel approach that simplifies graph structures through graph coarsening while preserving essential spatiotemporal dependencies. This coarsening process minimizes communication overhead and enables effective cross-subgraph message passing, capturing both local and long-range dependencies. ComPact further leverages hierarchical graph learning and structured edge collaboration to integrate global information into local subgraphs, enhancing predictive performance. Experimental validation on large-scale datasets, primarily the WindPower dataset, demonstrates ComPact's superiority in wind speed forecasting, with up to a 31.82% reduction in Mean Absolute Error (MAE) and 11.8% lower in Mean Absolute Percentage Error (MAPE) compared to federated learning baselines.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2320-2341"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979814","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888412","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010231
Zhi Pei;Chunlei Hong;Fen Xia;Chao Wang
Quantum computing is generally considered non-threatening to symmetric ciphers. Quantum attacks on symmetric ciphers require a thorough analysis of their internal structures, posing considerable difficulties and challenges. As of 2023, Google's quantum supremacy chip, Sycamore, is still incapable of cryptanalysis. Leveraging D-Wave's quantum annealing exploits the unique quantum tunneling effect, providing an edge in solving combinatorial optimization problems. It can be regarded as a class of artificial intelligence algorithm that can achieve global optimization. We propose a quantum heuristic symmetric cipher attack algorithm for substitution-permutation network (SPN) symmetric ciphers, which transforms the plaintext-ciphertext propagation rules within SPN structure into the problem of solving a constrained quadratic model (CQM). A novel reduction algorithm is employed to eliminate redundant constraint conditions. The D-Wave Advantage quantum computer is used to recover the encryption sub-keys. Using the quantum approximate optimization algorithm, IBM Q Experience can only recover two rounds of the Heys Cipher sub-key, whereas D-Wave Advantage achieves complete key recovery, validating its potential in quantum symmetric cipher attacks.
{"title":"An Innovative Algorithm for Attacking Symmetric Ciphers Using D-Wave Quantum Annealing","authors":"Zhi Pei;Chunlei Hong;Fen Xia;Chao Wang","doi":"10.26599/TST.2024.9010231","DOIUrl":"https://doi.org/10.26599/TST.2024.9010231","url":null,"abstract":"Quantum computing is generally considered non-threatening to symmetric ciphers. Quantum attacks on symmetric ciphers require a thorough analysis of their internal structures, posing considerable difficulties and challenges. As of 2023, Google's quantum supremacy chip, Sycamore, is still incapable of cryptanalysis. Leveraging D-Wave's quantum annealing exploits the unique quantum tunneling effect, providing an edge in solving combinatorial optimization problems. It can be regarded as a class of artificial intelligence algorithm that can achieve global optimization. We propose a quantum heuristic symmetric cipher attack algorithm for substitution-permutation network (SPN) symmetric ciphers, which transforms the plaintext-ciphertext propagation rules within SPN structure into the problem of solving a constrained quadratic model (CQM). A novel reduction algorithm is employed to eliminate redundant constraint conditions. The D-Wave Advantage quantum computer is used to recover the encryption sub-keys. Using the quantum approximate optimization algorithm, IBM Q Experience can only recover two rounds of the Heys Cipher sub-key, whereas D-Wave Advantage achieves complete key recovery, validating its potential in quantum symmetric cipher attacks.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2184-2194"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979788","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888336","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010102
Difeng Wang;Liangming Chen;Fang Gong;Qiankun Zhu
Recently, Randomly Wired Neural Networks (RWNNs) using random graphs for Convolutional Neural Network (CNN) construction have shown efficient layer connectivity, but may limit depth, affecting approximation, generalization, and robustness. In this work, we increase the depth of graph-structured CNNs while maintaining efficient pathway usage, which is achieved by building a feature-extraction backbone with a depth-first search, employing edges that have not been traversed for parameter-efficient skip connections. The proposed Efficiently Pathed Deep Network (EPDN) reaches maximum graph-based architecture depth without redundant node use, ensuring feature propagation with reduced connectivity. The deep structure of EPDN, coupled with its efficient pathway usage, allows for a nuanced feature extraction. EPDN is highly beneficial for processing remote sensing images, as its performance relies on the ability to resolve intricate spatial details. EPDN facilitates this by preserving low-level details through its deep and efficient skip connections, allowing for enhanced feature extraction. Additionally, the remote-sensing-adapted EPDN variant is akin to a special case of a multistep method for solving an Ordinary Differential Equation (ODE), leveraging historical data for improved prediction. EPDN outperforms existing CNNs in generalization and robustness on image classification benchmarks and remote sensing tasks. The source code is publicly available at https://github.com/AnonymousGithubLink/EPDN.
{"title":"Maximizing Depth of Graph-Structured Convolutional Neural Networks with Efficient Pathway Usage for Remote Sensing","authors":"Difeng Wang;Liangming Chen;Fang Gong;Qiankun Zhu","doi":"10.26599/TST.2024.9010102","DOIUrl":"https://doi.org/10.26599/TST.2024.9010102","url":null,"abstract":"Recently, Randomly Wired Neural Networks (RWNNs) using random graphs for Convolutional Neural Network (CNN) construction have shown efficient layer connectivity, but may limit depth, affecting approximation, generalization, and robustness. In this work, we increase the depth of graph-structured CNNs while maintaining efficient pathway usage, which is achieved by building a feature-extraction backbone with a depth-first search, employing edges that have not been traversed for parameter-efficient skip connections. The proposed Efficiently Pathed Deep Network (EPDN) reaches maximum graph-based architecture depth without redundant node use, ensuring feature propagation with reduced connectivity. The deep structure of EPDN, coupled with its efficient pathway usage, allows for a nuanced feature extraction. EPDN is highly beneficial for processing remote sensing images, as its performance relies on the ability to resolve intricate spatial details. EPDN facilitates this by preserving low-level details through its deep and efficient skip connections, allowing for enhanced feature extraction. Additionally, the remote-sensing-adapted EPDN variant is akin to a special case of a multistep method for solving an Ordinary Differential Equation (ODE), leveraging historical data for improved prediction. EPDN outperforms existing CNNs in generalization and robustness on image classification benchmarks and remote sensing tasks. The source code is publicly available at https://github.com/AnonymousGithubLink/EPDN.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"1940-1953"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979817","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888335","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-29DOI: 10.26599/TST.2024.9010257
Botao Liu;Ming Tang
The rapid proliferation of Internet of Things (IoT) devices necessitates lightweight cryptographic algorithms and their secure physical implementations. Masking, as a provably secure countermeasure against Side-Channel Attacks (SCA), has been extensively studied in the context of lightweight cryptography algorithms. Currently, some cryptographers have proposed a low-cost Threshold Implementation (TI) of the uBlock algorithm. However, their approach suffers from significant area overhead due to the inefficient serial and pipelined implementation of uBlock's Pshufb-Xor (PX) network structure. To address this issue, we develop a new serial and pipelined implementation method that optimizes the area of the uBlock algorithm. Based on this optimization, we implement a 2-share TI scheme for uBlock that requires minimal area resources and does not need fresh randomness. Compared to the state-of-the-art appoach, our method reduces slice area by 63.4% on Field Programmable Gate Arrays (FPGA) platform and Gate Equivalent (GE) area by 17.2% on Application-Specific Integrated Circuit (ASIC) platform for the unprotected implementation. For the protected implementation, our method reduces slice area by 41.5% and GE area by 14.0%. Finally, our protection scheme is validated using the automated tool PROLEAD and evaluated with Test Vector Leakage Assessment (TVLA), achieving first-order glitch-extended probing security.
{"title":"A Very Compact and a Threshold Implementation of uBlock for Internet of Things","authors":"Botao Liu;Ming Tang","doi":"10.26599/TST.2024.9010257","DOIUrl":"https://doi.org/10.26599/TST.2024.9010257","url":null,"abstract":"The rapid proliferation of Internet of Things (IoT) devices necessitates lightweight cryptographic algorithms and their secure physical implementations. Masking, as a provably secure countermeasure against Side-Channel Attacks (SCA), has been extensively studied in the context of lightweight cryptography algorithms. Currently, some cryptographers have proposed a low-cost Threshold Implementation (TI) of the uBlock algorithm. However, their approach suffers from significant area overhead due to the inefficient serial and pipelined implementation of uBlock's Pshufb-Xor (PX) network structure. To address this issue, we develop a new serial and pipelined implementation method that optimizes the area of the uBlock algorithm. Based on this optimization, we implement a 2-share TI scheme for uBlock that requires minimal area resources and does not need fresh randomness. Compared to the state-of-the-art appoach, our method reduces slice area by 63.4% on Field Programmable Gate Arrays (FPGA) platform and Gate Equivalent (GE) area by 17.2% on Application-Specific Integrated Circuit (ASIC) platform for the unprotected implementation. For the protected implementation, our method reduces slice area by 41.5% and GE area by 14.0%. Finally, our protection scheme is validated using the automated tool PROLEAD and evaluated with Test Vector Leakage Assessment (TVLA), achieving first-order glitch-extended probing security.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 5","pages":"2270-2283"},"PeriodicalIF":6.6,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10979790","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888377","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：