Article
Engineering, Electrical & Electronic
Tian Yuan, Weiqiang Liu, Jie Han, Fabrizio Lombardi
Summary: This paper introduces a hardware-oriented CNN compression strategy, which combines layers with and without pruning to achieve a balance between compression ratio and processing efficiency. Through hardware/algorithm co-optimization, a NP-P hybrid compressed CNN model was successfully implemented on FPGAs.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2021)
Article
Computer Science, Artificial Intelligence
Linyong Huang, Zhe Zhang, Zhaoyang Du, Shuangchen Li, Hongzhong Zheng, Yuan Xie, Nianxiong Tan
Summary: This paper introduces the application of Product Quantization (PQ) on Graph Neural Networks (GNNs) to achieve superior memory capacity reduction. By proposing Enhanced Product Quantization (EPQ) and an efficient quantization framework, GNNs can be deployed on resource-constrained devices, enabling improved compression performance and computation acceleration.
Article
Computer Science, Artificial Intelligence
Kaixuan Yao, Feilong Cao, Yee Leung, Jiye Liang
Summary: This paper proposes a method to compress deep neural networks based on interpretability, achieving effective compression while providing a better interpretation of the deep learning process. By utilizing single-layer filter pruning, the entire DNN model can be compressed layer by layer, resulting in reduced computation cost and storage requirements for implementing complex DNN models in small mobile devices.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Sheng Xu, Hanlin Chen, Xuan Gong, Kexin Liu, Jinhu Lu, Baochang Zhang
Summary: The paper introduces an efficient end-to-end pruning method based on feature stabilization (EPFS) for structured pruning, which incorporates Center Loss and fast iterative shrinkage-thresholding algorithm (FISTA) to improve pruning efficiency and accuracy.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Engineering, Electrical & Electronic
Xuepeng Chang, Huihui Pan, Weiyang Lin, Huijun Gao
Summary: The paper proposes a framework containing model compression and hardware acceleration to address the deployment of CNN on embedded devices, including mixed pruning method, data storage optimization, and FPGA accelerator.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2021)
Article
Engineering, Electrical & Electronic
Jun Shi, Jianfeng Xu, Kazuyuki Tasaka, Zhibo Chen
Summary: This paper proposes a SASL approach for further optimization, utilizing saliency estimation and adjusting regularization strength according to saliency to better preserve prediction performance. A hard sample mining strategy is utilized during the pruning phase to optimize the data-dependent criterion, showing higher effectiveness and efficiency.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Artificial Intelligence
Wenxiao Wang, Zhengxu Yu, Cong Fu, Deng Cai, Xiaofei He
Summary: As deep CNNs become larger, deploying them on mobile devices with limited resources becomes more challenging. Filter-level pruning is a popular method to compress deep models for mobile deployment, but it still faces issues such as high redundancy and sub-optimality.
Article
Computer Science, Artificial Intelligence
Peng Peng, Mingyu You, Weisheng Xu, Jiaxin Li
Summary: The paper presents a novel quantization approach to efficiently deploy deep convolutional neural networks on mobile devices, solving the issue of datatype mismatch in low bitwidth quantization methods. By improving the quantization function and simultaneously quantizing batch normalization parameters, the proposed method achieves comparable prediction accuracy with reduced run-time latency while maintaining the advantages of low bitwidth representation.
Review
Computer Science, Artificial Intelligence
Chunyu Yuan, Sos S. Agaian
Summary: Deep learning has revolutionized the development of intelligent systems and is widely used in various real-life applications. Binary Neural Networks (BNN) are game-changing technologies that can enhance the capabilities of deep learning and can be implemented on computationally limited and energy-constrained devices. While BNNs can significantly save storage, computation cost, and energy consumption, they often trade-offs with extra memory, computation cost, and higher performance.
ARTIFICIAL INTELLIGENCE REVIEW
(2023)
Article
Computer Science, Artificial Intelligence
Tailin Liang, John Glossner, Lei Wang, Shaobo Shi, Xiaotong Zhang
Summary: Deep neural networks have been widely used in computer vision applications, but their complex architectures pose challenges in real-time deployment due to high computation resources and energy costs. Network compression techniques such as pruning and quantization can help overcome these challenges by reducing redundant computations. Both techniques can be used independently or together to improve the efficiency and performance of deep neural networks.
Article
Engineering, Electrical & Electronic
Jiawei Xu, Yuxiang Huan, Boming Huang, Haoming Chu, Yi Jin, Li-Rong Zheng, Zhuo Zou
Summary: The study introduces a memory-efficient CNN accelerator design for resource-constrained devices, achieving model compression and precision improvement through SegLog quantization. The ASIC and FPGA implementations demonstrate high area efficiency and memory efficiency, respectively.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2021)
Article
Computer Science, Information Systems
Dheeraj Kumar, Mayuri A. Mehta, Vivek C. Joshi, Rachana S. Oza, Ketan Kotecha, Jerry Chun-Wei Lin
Summary: This research proposes a detailed classification of model acceleration methods, including a broad classification of filter pruning methods. It then conducts an empirical evaluation of four filter pruning methods and demonstrates the effect of filter pruning on both pre-trained CNN and custom CNN through experimental results.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Environmental Sciences
Jiaqing Zhang, Jie Lei, Weiying Xie, Daixun Li
Summary: The fusion of hyperspectral and LiDAR images is crucial for accurate classification and recognition in remote sensing. This study proposes a method that combines a binary convolutional neural network and a graph convolutional network with invariant attributes to overcome the challenges of constructing effective graph structures. The method utilizes a joint detection framework to simultaneously learn features from regular and irregular regions, resulting in an enhanced structural representation of the images. Experimental results demonstrate the superior performance of the proposed method in hyperspectral image analysis tasks.
Article
Computer Science, Artificial Intelligence
Babak Rokh, Ali Azarpeyvand, Alireza Khanteymoori
Summary: Recent advancements in machine learning achieved by Deep Neural Networks (DNNs) have led to significant improvements in accuracy; however, the high number of parameters and computations associated with DNNs result in high memory usage and energy consumption. To address this issue, various compression techniques have been employed, with quantization being a promising approach. This article presents a comprehensive survey of quantization concepts and methods, focusing on image classification. It covers clustering-based quantization methods, the use of scale factor parameter, training of a quantized DNN, replacement of floating-point operations with bitwise operations, sensitivity of different layers in quantization, evaluation metrics, and benchmarks. The article aims to familiarize readers with quantization concepts, introduce important works in DNN quantization, and highlight challenges for future research in this field.
ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY
(2023)
Article
Biochemical Research Methods
Yuan Li, Xu Shi, Liping Yang, Chunyu Pu, Qijuan Tan, Zhengchun Yang, Hong Huang
Summary: This paper proposes a multi-layer collaborative generative adversarial transformer (MC-GAT) for cholangiocarcinoma (CCA) classification from hyperspectral pathological images. MC-GAT consists of a generator and a discriminator, which improve the model's generalization and discriminating power. Experimental results show that MC-GAT achieves better classification results compared to other methods.
BIOMEDICAL OPTICS EXPRESS
(2022)
Article
Oncology
Yueming Wang, Jiwei Huang, Cuijian Zhang, Xiaoyi Hu, Ping Wang, Guohai Shi, Jin Zhang, Wen Kong, Yonghui Chen, Yiran Huang, Dingwei Ye, Dan Xia, Jianming Guo, Zhisong He, Wei Xue
Summary: This retrospective multicenter study analyzed the clinical outcomes and prognostic factors of 85 patients who underwent targeted therapy for local retroperitoneal recurrence (RPR) after radical nephrectomy (RN). The study found that presurgical targeted therapy followed by surgical resection was associated with better cancer-specific survival (CSS) compared to targeted therapy alone. Factors such as risk classification, number of recurrence lesions, and surgical resection were independent predictors of CSS.
INTERNATIONAL JOURNAL OF CANCER
(2023)
Article
Automation & Control Systems
Hongyuan Wang, Jingcheng Wang, Haotian Xu, Shangwei Zhao
Summary: This paper proposes a self-triggered stochastic model predictive control (SMPC) algorithm for networked control systems with disturbances and communication frequency restrictions. The algorithm integrates the design of control inputs and sampling interval to reduce communication while ensuring performance. Chance constraints are transformed into deterministic ones, and the nonconvex terms in covariance propagation are avoided using an upper bound control approach. The proposed algorithm is shown to be effective and achieves the desired performance through numerical simulations.
INTERNATIONAL JOURNAL OF CONTROL
(2023)
Article
Chemistry, Physical
Yihui Li, Deying Lin, Yongfu Li, Peikun Jiang, Xiaobo Fang, Bing Yu
Summary: In this study, bimetallic Fe/Mn-loaded hydroxyl-rich biochar (FeMn-OH-BC) is synthesized to activate peroxymonosulfate (PMS) through nonradical-dominated pathways for degrading organic pollutants. The FeMn-OH-BC exhibits excellent catalytic activity at a wide range of pH conditions and is not inhibited by various anions. Electron paramagnetic resonance measurements and quenching tests confirm that O-1(2) is the major reactive oxygen species generated from FeMn-OH-BC based PMS activation.
Article
Dermatology
Ruzhen Luo, Yunan Ji, Yan-hui Liu, Hongyu Sun, Siyuan Tang, Xuechun Li
Summary: This study examines the relationships between social support, decision regret, self-stigma, and quality of life in patients with diabetic foot ulcers. The results show that quality of life is negatively correlated with self-stigma, positively correlated with social support and giving up coping, and not significantly correlated with confrontation coping and avoidance coping.
INTERNATIONAL WOUND JOURNAL
(2023)
Article
Clinical Neurology
Dongling Zhang, Liche Zhou, Yuting Shi, Jun Liu, Hongjiang Wei, Qiqi Tong, Hongjian He, Tao Wu
Summary: The study found that the free water (FW) values in the posterior substantia nigra (pSN) were significantly elevated and continued to increase in asymptomatic LRRK2 G2019S mutation carriers. There was also a negative correlation between FW changes in the left pSN and striatal binding ratio (SBR) changes in the left putamen.
MOVEMENT DISORDERS
(2023)
Article
Optics
Ling-Dong Kong, Hui Wang, Qing-Yuan Zhao, Jia-Wei Guo, Yang-Hui Huang, Hao Hao, Shi Chen, Xue-Cou Tu, La-Bao Zhang, Xiao-Qing Jia, Lin Kang, Jian Chen, Pei-Heng Wu
Summary: A kilopixel imager is developed by introducing an orthogonal time-amplitude-multiplexing method in superconducting nanowire single-photon detectors. The readout is solely built in the nanowire structure to manipulate the hotspot growth and microwave propagation after photon detection. The experiment results show that the imager has high pixel fidelity and temporal resolution, making it suitable for single-photon imaging experiments in photon-starved conditions. This research is of great importance for the development of large-scale single-photon imagers in fields such as quantum measurements, remote sensing, and astronomical telescopes.
Article
Public, Environmental & Occupational Health
Lina Huang, Liusen Wang, Hongru Jiang, Huijun Wang, Zhihong Wang, Bing Zhang, Gangqiang Ding
Summary: There have been significant changes in the macronutrient composition of the Chinese adult diet over the past few decades, with a decrease in the percentage of energy consumed from carbohydrates and an increase in the percentage of energy consumed from total fat. The diet quality remains suboptimal.
PUBLIC HEALTH NUTRITION
(2023)
Article
Environmental Sciences
Jingli Yang, Yongbin Lu, Yana Bai, Zhiyuan Cheng
Summary: Growing studies have shown a potential link between metal exposure and diabetes risk. This study used a multiple linear regression model to examine the associations between urinary metals (cobalt and molybdenum) and diabetes-related indicators in a cross-sectional study. The results demonstrated significant sex-specific and dose-response relationships between urinary metals and diabetes-related indicators, suggesting the need for further investigation into the underlying mechanisms.
JOURNAL OF ENVIRONMENTAL SCIENCES
(2023)
Article
Plant Sciences
Liqian Chen, Xinghong Zhou, Yijian Deng, Ying Yang, Xiaohu Chen, Qinghong Chen, Yanyan Liu, Xiuqiong Fu, Hiu Yee Kwan, Yanting You, Wen Jin, Xiaoshan Zhao
Summary: This study aimed to confirm the protective effects of ZWD on cardiac hypertrophy and explore the underlying mechanisms. The results showed that ZWD reduces oxidative stress and inflammation and exerts cardioprotective effects by activating the sGC-cGMP-PKG signaling pathway.
JOURNAL OF ETHNOPHARMACOLOGY
(2023)
Article
Genetics & Heredity
Jiaoyang Chen, Yi Chen, Ying Yang, Xueyang Niu, Jing Zhang, Qi Zeng, Aijie Liu, Xiaojing Xu, Xiaoxu Yang, Shupin Li, Xiaoling Yang, Yi Wang, Yuehua Zhang
Summary: This study investigated the occurrence of mosaicism in epilepsy probands and their parents, and found that mosaic phenomenon is not rare in families with epilepsy. The phenotypes of mosaic parents were milder or normal, and ADS is a reliable way of mosaicism detection for clinical application.
JOURNAL OF HUMAN GENETICS
(2023)
Article
Computer Science, Artificial Intelligence
Tianli Zhao, Qinghao Hu, Xiangyu He, Weixiang Xu, Jiaxing Wang, Cong Leng, Jian Cheng
Summary: This article introduces an efficient convolution algorithm called ECBC, which achieves high computation performance by utilizing highly optimized subroutines in matrix multiplication and dividing the convolution computation into blocks using a block algorithm. This approach helps reduce the memory overhead significantly.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Pharmacology & Pharmacy
Xiaoxi Lv, Shanshan Liu, Chang Liu, Yunxuan Li, Tingting Zhang, Jie Qi, Ke Li, Fang Hua, Bing Cui, Xiaowei Zhang, Yuxin Liu, Jiaojiao Yu, Jinmei Yu, Li Li, Xia Li, Zhigang Yao, Bo Huang
Summary: This study reveals that repetitive lung damage leads to the accumulation of the transcriptional repressor SLUG, inhibiting lung alveolar regeneration and causing the failure of pulmonary fibrosis. The elevated SLUG represses the expression of phosphate transporter SLC34A2, reducing intracellular phosphate and inhibiting key kinases JNK and P38 MAPK, resulting in LAR failure. Targeting the TRIB3/MDM2 interaction can restore LAR capacity and provide potential therapeutic efficacy against fibroproliferative lung diseases.
ACTA PHARMACEUTICA SINICA B
(2023)
Article
Engineering, Biomedical
Hongwei Wu, Yuna Shang, Wei Sun, Xinyi Ouyang, Wenyan Zhou, Jieji Lu, Shuhui Yang, Wei Wei, Xudong Yao, Xiaozhao Wang, Xianzhu Zhang, Yishan Chen, Qiulin He, Zhimou Yang, Hongwei Ouyang
Summary: This study developed a strategy to enhance integration of the gap region following mosaicplasty by applying BSN-GelMA hydrogel. The results showed that BSN-GelMA achieved seamless osteochondral healing in the gap region, with improved cartilage repair scores, glycosaminoglycan content, subchondral bone volume, and collagen II expression. This study provides a powerful approach to improve gap integration after autologous mosaicplasty and offers a promising bioactive material for cell-free tissue regeneration.
BIOACTIVE MATERIALS
(2023)
Article
Computer Science, Artificial Intelligence
Peisong Wang, Fanrong Li, Gang Li, Jian Cheng
Summary: In this article, the advantages of extremely sparse networks with binary connections for image classification through software-hardware codesign are investigated. A binary augmented extremely pruning method is proposed to achieve high sparsity with minimal accuracy degradation, and a hardware architecture based on the resulting sparse and binary networks is designed to explore the benefits of extreme sparsity with negligible consumption. Experiments on ImageNet classification and FPGA demonstrate a significant tradeoff between accuracy and efficiency in the proposed software-hardware architecture.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Bo Jiang, Leiling Wang, Jian Cheng, Jin Tang, Bin Luo
Summary: Compact representation of graph data is a fundamental problem in pattern recognition and machine learning. Graph neural networks (GNNs) have been widely studied for graph-structured data representation and learning. In this article, a new model called graph propagation-embedding networks (GPENs) is proposed, which integrates feature propagation and low-dimensional embedding into a unified network. GPENs have good motivation and can be well-explained, and they guarantee both compactness and efficiency. Experiments demonstrate the effectiveness and benefits of GPENs and M-GPENs.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)