检索结果-内蒙古大学图书馆

DMRIntTk: Integrating different DMR sets based on density peak clustering

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhang, Wenjin Jie, Wenlong Cui, Wanxin Duan, Guihua Zou, You Peng, Xiaoqing Center for Medical Genetics Hunan Key Laboratory of Medical Genetics School of Life Sciences Central South University Changsha410083 China Hunan Key Laboratory of Bioinformatics School of Computer Science and Engineering Central South University Changsha410083 China High Performance Computing Center Central South University Changsha410083 China

Background: Identifying differentially methylated regions (DMRs) is a basic task in DNA methylation analysis. However, due to the different strategies adopted, different DMR sets will be predicted on the same dataset, which poses a challenge in selecting a reliable and comprehensive DMR set for downstream analysis. Results: Here, we develop DMRIntTk, a toolkit for integrating DMR sets predicted by different methods on a same dataset. In DMRIntTk, the genome is segmented into bins and the reliability of each DMR set at different methylation thresholds is evaluated. Then, the bins are weighted based on the covered DMR sets and integrated into DMRs by using a density peak clustering algorithm. To demonstrate the practicality of DMRIntTk, DMRIntTk was applied to different scenarios, including different tissues with relatively large methylation differences, cancer tissues versus normal tissues with medium methylation differences, and disease tissues versus normal tissues with subtle methylation differences. The results show that DMRIntTk can effectively trim the regions with small methylation differences in the original DMR sets and therefore it can enhance the proportion of DMRs with higher methylation differences. In addition, the overlap analysis suggests that the integrated DMR sets are quite comprehensive, and the functional analysis indicates the integrated disease-related DMR sets are significantly enriched in biological pathways, which are associated with the pathological mechanisms of the diseases. Conclusions: Conclusively, DMRIntTk can help researchers obtaining a reliable and comprehensive DMR set from many prediction methods. © 2024, CC BY.

关键词： Diseases

HPCR: Holistic Proxy-based Contrastive Replay for Online Continual Learning

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Lin, Huiwei Feng, Shanshan Zhang, Baoquan Li, Xutao Ye, Yunming Depart ment of Computer Science Harbin Institute of Technology Shenzhen518055 China Shenzhen Key Laboratory of Internet Informatio Collaboration Shenzhen518055 China Centre for Frontier AI Research Insti tute of High Performance Computing A*STAR Singapore

Online continual learning, aimed at developing a neural network that continuously learns new data from a single pass over an online data stream, generally suffers from catastrophic forgetting. Existing replay-based methods alleviate forgetting by replaying partial old data in a proxy-based or contrastive-based replay manner, each with its own shortcomings. Our previous work proposes a novel replay-based method called proxy-based contrastive replay (PCR), which handles the shortcomings by achieving complementary advantages of both replay manners. In this work, we further conduct gradient and limitation analysis of PCR. The analysis results show that PCR still can be further improved in feature extraction, generalization, and anti-forgetting capabilities of the model. Hence, we develop a more advanced method named holistic proxy-based contrastive replay (HPCR). HPCR consists of three components, each tackling one of the limitations of PCR. The contrastive component conditionally incorporates anchor-to-sample pairs to PCR, improving the feature extraction ability. The second is a temperature component that decouples the temperature coefficient into two parts based on their gradient impacts and sets different values for them to enhance the generalization ability. The third is a distillation component that constrains the learning process with additional loss terms to improve the anti-forgetting ability. Experiments on four datasets consistently demonstrate the superiority of HPCR over various state-of-the-art methods. Copyright © 2023, The Authors. All rights reserved.

关键词： Contrastive Learning

Dynamic recrystallization-induced temperature insensitivity of yield stress in single-crystal Al1.2CrFeCoNi micropillars

学校读者我要写书评

暂无评论

science China(Technological sciences) 2021年第1期64卷 11-22页

作者： HUANG RuiRui ZHANG Qian ZHANG Xuan LI JianGuo CAO TangQing YAO JiaHao XUE YunFei GAO HuaJian LI XiaoYan Center for Advanced Mechanics and Materials Applied Mechanics LaboratoryDepartment of Engineering MechanicsTsinghua UniversityBeijing 100084China School of Materials Science and Engineering Beijing Institute of TechnologyBeijing 100081China Shenyang National Laboratory for Materials Science Institute of Metal ResearchChinese Academy of SciencesShenyang 110016China School of Mechanical and Aerospace Engineering College of EngineeringNanyang Technological UniversitySingapore 639798Singapore Institute of High Performance Computing A*STARSingapore 138632Singapore

high-entropy alloys, a new class of metallic materials, exhibit excellent mechanical properties at high temperatures. In spite of the worldwide interest, the underlying mechanisms for temperature dependence of mechanical properties of these alloys remain poorly understood. Here, we systemically investigate the mechanical behaviors and properties of Al_(1.2)CrFeCoNi(comprising a body-centered cubic phase) and Al_(0.3)CrFeCoNi(comprising a face-centered cubic phase) single-crystal micropillars with three orientations([100], [110], and [111]) at temperatures varying from 300 to 675 K by using in situ compression of micropillars inside a scanning electron microscope. The results show that the yield stresses of Al_(1.2)CrFeCoNi micropillars are insensitive to temperature changes, and their flow stresses and work hardening rates increase slightly with increasing temperature from 300 to550 K, which differs from the typical temperature dependence of yield/flow stresses in metals and alloys. In contrast,Al_(0.3)CrFeCoNi micropillars exhibit typical thermal softening. Furthermore, it is found that the Al_(1.2)CrFeCoNi micropillars exhibit a transition from homogenous deformation to localized deformation at a critical temperature, while the Al_(0.3)CrFeCoNi micropillars always maintain a well-distributed and fine slip deformation. Detailed transmission electron microscopy analyses reveal that dynamic recrystallization(involving dislocation tangles, and formation of dislocation cell structures and sub-grains)plays a key role in the observed temperature insensitivity of the yield stress and increasing flow stress(and work hardening rate)with increasing temperature in the Al_(1.2)CrFeCoNi micropillars, and that thermally activated dislocation slip leads to thermal softening of the Al_(0.3)CrFeCoNi micropillars. The differences in deformation modes and temperature dependence of the mechanical properties between Al_(1.2)CrFeCoNi and Al_(0.3)CrFeCoNi essentially originate from the d

关键词： high entropy alloy temperature insensitivity dynamic recrystallization micropillar deformation twin

DESIGN AND DEVELOPMENT OF MORPHOLOGICAL ANALYZER FOR TIGRIGNA VERBS USING HYBRID APPROACH

学校读者我要写书评

暂无评论

Research Square

Research Square 2023年

作者： Gebremeskel, Hagos Gebremedhin Chong, Feng Heyan, Huang Beijing Engineering Research Center of High Volume Language Information Processing Cloud Computing Applications Laboratory School of Computer Science and Technology Beijing Institute of Technology China Department of Computer Science Mekelle University Tigray Mekelle231 Ethiopia

Morphological analyzer is the base for various high-level NLP applications such as information retrieval, spell checking, grammar checking, machine translation, speech recognition, POS tagging and automatic sentence construction. This paper is carefully designed for design and analysis of morphological analyzer Tigrigna verbs using hybrid of memory learning and rules based approaches. The experiment have conducted using Python 3 where TiMBL algorithms IB2 and TRIBL2, and Finite State Transducer rules are used. The performance of the system has been evaluated using 10 fold cross validation technique. Testing was conducted using optimized parameter settings for regular verbs and linguistic rules of the Tigrigna language allomorph and phonology for the irregular verbs. The accuracy of the memory based approach with optimized parameters of TiMBL algorithm IB2 and TRIBL2 was 93.24% and 92.31%, respectively. Finally, the hybrid approach had an actual performance of 95.6% using linguistic rules for handling irregular and copula verbs. © 2023, CC BY.

关键词： Machine learning

Progressive Perception Learning for Distribution Modulation in Siamese Tracking

学校读者我要写书评

暂无评论

Progressive Perception Learning for Distribution Modulation ...

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Kun Hu Xianchen Zhou Mingyu Cao Mengzhu Wang Guangjie Gao Wenjing Yang Huibin Tan State Key Laboratory of High Performance Computing National University of Defense Technology Changsha China College of Sciences National University of Defense Technology Changsha China College of Computer Science and Technology National University of Defense Technology Changsha China

We explore an innovative view on distribution modulation to boost Siamese trackers. Specially, we observed two cases of possible distribution inconsistency in Siamese tracking: 1) Two branches with different sizes may be in different distribution ranges after a shared backbone (including BN layers). 2) The background data may affect the total feature distribution of the search branch. To address these issues, we proposed a plug-and-play component named Progressive Perception Learning Module (P 2 LM) to modulate the distribution using three feature normalization blocks successively, i.e., Self-Aware Block (SAB), Target-Aware Block (TAB), and Region-Aware Block (RAB). SAB regulates the distribution of each branch independently for the first issue. TAB uses the target information to guide the distribution adjustments of the two branches. RAB divides the search image into foreground and background with a region mask and normalizes them separately to filter the background distractors for robust tracking. TAB and RAB synergistically alleviate the distribution shifts caused by environmental variance. Experiments on OTB100, UAV123, LaSOT, and GOT-10k verify the compelling effects of our module.

关键词： Target tracking Modulation Information filters Acoustics Speech processing Self-aware

Verifying Safety of Neural Networks from Topological Perspectives

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Liang, Zhen Ren, Dejin Xue, Bai Wang, Ji Yang, Wenjing Liu, Wanwei National University of Defense Technology Institute for Quantum Information State Key Laboratory of High Performance Computing Hunan Changsha410073 China Chinese Academy of Sciences State Key Lab. of Computer Science BeijingBAI XUE China

Neural networks (NNs) are increasingly applied in safety-critical systems such as autonomous vehicles. However, they are fragile and are often ill-behaved. Consequently, their behaviors should undergo rigorous guarantees before deployment in practice. In this paper, we propose a set-boundary reachability method to investigate the safety verification problem of NNs from a topological perspective. Given an NN with an input set and a safe set, the safety verification problem is to determine whether all outputs of the NN resulting from the input set fall within the safe set. In our method, the homeomorphism property and the open map property of NNs are mainly exploited, which establish rigorous guarantees between the boundaries of the input set and the boundaries of the output set. The exploitation of these two properties facilitates reachability computations via extracting subsets of the input set rather than the entire input set, thus controlling the wrapping effect in reachability analysis and facilitating the reduction of computation burdens for safety verification. The homeomorphism property exists in some widely used NNs such as invertible residual networks (i-ResNets) and Neural ordinary differential equations (Neural ODEs), and the open map is a less strict property and easier to satisfy compared with the homeomorphism property. For NNs establishing either of these properties, our set-boundary reachability method only needs to perform reachability analysis on the boundary of the input set. Moreover, for NNs that do not feature these properties with respect to the input set, we explore subsets of the input set for establishing the local homeomorphism property and then abandon these subsets for reachability computations. Finally, some examples demonstrate the performance of the proposed *** Codes 68Q60, 68T07 Copyright © 2023, The Authors. All rights reserved.

关键词： Topology

Repairing Deep Neural Networks Based on Behavior Imitation

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Liang, Zhen Wu, Taoran Zhao, Changyuan Liu, Wanwei Xue, Bai Yang, Wenjing Wang, Ji Institute for Quantum Information State Key Laboratory of High Performance Computing National University of Defense Technology Changsha China State Key Laboratory of Computer Science Institute of Software Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Beijing China College of Computer Science and Technology National University of Defense Technology Changsha China

The increasing use of deep neural networks (DNNs) in safety-critical systems has raised concerns about their potential for exhibiting ill-behaviors. While DNN verification and testing provide post hoc conclusions regarding unexpected behaviors, they do not prevent the erroneous behaviors from occurring. To address this issue, DNN repair/patch aims to eliminate unexpected predictions generated by defective DNNs. Two typical DNN repair paradigms are retraining and fine-tuning. However, existing methods focus on the high-level abstract interpretation or inference of state spaces, ignoring the underlying neurons' outputs. This renders patch processes computationally prohibitive and limited to piecewise linear (PWL) activation functions to great extent. To address these shortcomings, we propose a behavior-imitation based repair framework, BIRDNN, which integrates the two repair paradigms for the first time. BIRDNN corrects incorrect predictions of negative samples by imitating the closest expected behaviors of positive samples during the retraining repair procedure. For the fine-tuning repair process, BIRDNN analyzes the behavior differences of neurons on positive and negative samples to identify the most responsible neurons for the erroneous behaviors. To tackle more challenging domain-wise repair problems (DRPs), we synthesize BIRDNN with a domain behavior characterization technique to repair buggy DNNs in a probably approximated correct style. We also implement a prototype tool based on BIRDNN and evaluate it on ACAS Xu DNNs. Our experimental results show that BIRDNN can successfully repair buggy DNNs with significantly higher efficiency than state-of-the-art repair tools. Additionally, BIRDNN is highly compatible with different activation *** Codes 68N99, , 68T99 Copyright © 2023, The Authors. All rights reserved.

关键词： Deep neural networks

Retention Leveling: Leverage Retention Refreshing and Wear Leveling Techniques to Enhance Flash Reliability with the Awareness of Temperature

学校读者我要写书评

暂无评论

Retention Leveling: Leverage Retention Refreshing and Wear L...

IEEE Non-Volatile Memory Systems and Applications Symposium (NVMSA)

作者： Wei-Chen Wang Chien-Chung Ho Yuan-Hao Chang Tei-Wei Kuo Yu-Ming Chang Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology Department of Computer Science and Information Engineering National Cheng Kung University Institute of Information Science Academia Sinica Depanment of Computer Science and Information Engineering National Taiwan University High Performance and Scientific Computing Center National Taiwan University Mohamed bin Zayed University of Artificial Intelligence Wolley Inc.

Wear leveling cannot entirely mitigate retention errors, which have recently become $a$ substantial reliability issue due to cell shrinking. This paper focuses on preserving accurate data by addressing retention errors rather than uniformly distributing erases across blocks, as maintaining accurate data is the true goal for enhancing reliability. We propose the concept of retention leveling, considering $a$ crucial factor, i.e., temperature, and compare it to traditional wear leveling. Experimental results demonstrate that retention leveling can significantly improve flash memory reliability by limiting the number of retention errors to merely 0.0019% of the conventional approach, while substantially alleviating the write amplification problem.

关键词：

Comparison of Multi-Phase Power Converters and Power Delivery Networks for Next-Generation Space Architectures

学校读者我要写书评

暂无评论

Comparison of Multi-Phase Power Converters and Power Deliver...

IEEE Conference on Aerospace

作者： Nicholas Franconi Thomas Cook Christopher Wilson Alan D. George NASA Goddard Space Flight Center Science Data Processing Branch Code 587 Software Engineering Division Department of Electrical and Computer Engineering NSF Center for Space High-Performance and Resilient Computing (SHREC) University of Pittsburgh

Next-generation spacecraft developers are earnestly investigating the application of artificial intelligence (AI) algorithms onboard to enable new mission concepts for space exploration and science. However, the current generation of radiation-hardened processors are inferior compared to commercial-off-the-shelf alternatives in terms of the computational performance required by modern AI applications. To address this disparity, space-system designers have started employing novel radiation-tolerant architectures combining both commercial and radiation-hardened components to mitigate radiation effects at a system level. Unfortunately, developing single-board computers with radiation-tolerant, high-performance processors is challenging because designers must balance the sparce selection of radiation-hardened power converters and high-reliability decoupling capacitors with SmallSat/CubeSat area constraints and limited thermal conduction. Consequently, the next-generation of space processors, including the AMD-Xilinx Versal Adaptive Compute Acceleration Platform require demanding power solutions capable of supplying core rails with 0.8V ± 18mV and currents up to 150 A. In this paper, we present a multifaceted analysis of the power system and decoupling network for a future Versal-based design. We developed preliminary power estimates based on expected processor and FPGA resource utilization for common AI processing applications. These estimates drive the power system requirements and a comparative analysis of single-phase integrated converters and multi-phase discrete converters for high-current FPGA supplies. We develop a tradespace between the number of phases, input voltage, load current, switching frequency, power efficiency, and printed circuit board area. Finally, we design and simulate four power delivery networks based on commercial, high-reliability, and flight-qualified capacitors to compare the efficacy of 0201 decoupling capacitors in flight missions.

关键词： Rails Program processors Capacitors Voltage Power systems Resource management Artificial intelligence