Search

Your search keyword '"Zhang, Pengtao"' showing total 167 results
Start Over Author "Zhang, Pengtao"
167 results on '"Zhang, Pengtao"'

1. MemoNet: Memorizing All Cross Features' Representations Efficiently via Multi-Hash Codebook Network for CTR Prediction

Author

Zhang, Pengtao and Zhang, Junlin

Subjects
Computer Science - Information Retrieval, Computer Science - Artificial Intelligence, Computer Science - Machine Learning
Abstract

New findings in natural language processing (NLP) demonstrate that the strong memorization capability contributes a lot to the success of Large Language Models (LLM). This inspires us to explicitly bring an independent memory mechanism into CTR ranking model to learn and memorize cross features' representations. In this paper, we propose multi-Hash Codebook NETwork (HCNet) as the memory mechanism for efficiently learning and memorizing representations of cross features in CTR tasks. HCNet uses a multi-hash codebook as the main memory place and the whole memory procedure consists of three phases: multi-hash addressing, memory restoring, and feature shrinking. We also propose a new CTR model named MemoNet which combines HCNet with a DNN backbone. Extensive experimental results on three public datasets and online test show that MemoNet reaches superior performance over state-of-the-art approaches. Besides, MemoNet shows scaling law of large language model in NLP, which means we can enlarge the size of the codebook in HCNet to sustainably obtain performance gains. Our work demonstrates the importance and feasibility of learning and memorizing representations of cross features, which sheds light on a new promising research direction.

Published
2022
Full Text
View/download PDF

2. FiBiNet++: Reducing Model Size by Low Rank Feature Interaction Layer for CTR Prediction

Author

Zhang, Pengtao, Zheng, Zheng, and Zhang, Junlin

Subjects
Computer Science - Information Retrieval, Computer Science - Artificial Intelligence
Abstract

Click-Through Rate (CTR) estimation has become one of the most fundamental tasks in many real-world applications and various deep models have been proposed. Some research has proved that FiBiNet is one of the best performance models and outperforms all other models on Avazu dataset. However, the large model size of FiBiNet hinders its wider application. In this paper, we propose a novel FiBiNet++ model to redesign FiBiNet's model structure, which greatly reduces model size while further improves its performance. One of the primary techniques involves our proposed "Low Rank Layer" focused on feature interaction, which serves as a crucial driver of achieving a superior compression ratio for models. Extensive experiments on three public datasets show that FiBiNet++ effectively reduces non-embedding model parameters of FiBiNet by 12x to 16x on three datasets. On the other hand, FiBiNet++ leads to significant performance improvements compared to state-of-the-art CTR methods, including FiBiNet.

Published
2022
Full Text
View/download PDF

5. ContextNet: A Click-Through Rate Prediction Framework Using Contextual information to Refine Feature Embedding

Author

Wang, Zhiqiang, She, Qingyun, Zhang, PengTao, and Zhang, Junlin

Subjects
Computer Science - Information Retrieval, Computer Science - Artificial Intelligence
Abstract

Click-through rate (CTR) estimation is a fundamental task in personalized advertising and recommender systems and it's important for ranking models to effectively capture complex high-order features.Inspired by the success of ELMO and Bert in NLP field, which dynamically refine word embedding according to the context sentence information where the word appears, we think it's also important to dynamically refine each feature's embedding layer by layer according to the context information contained in input instance in CTR estimation tasks. We can effectively capture the useful feature interactions for each feature in this way. In this paper, We propose a novel CTR Framework named ContextNet that implicitly models high-order feature interactions by dynamically refining each feature's embedding according to the input context. Specifically, ContextNet consists of two key components: contextual embedding module and ContextNet block. Contextual embedding module aggregates contextual information for each feature from input instance and ContextNet block maintains each feature's embedding layer by layer and dynamically refines its representation by merging contextual high-order interaction information into feature embedding. To make the framework specific, we also propose two models(ContextNet-PFFN and ContextNet-SFFN) under this framework by introducing linear contextual embedding network and two non-linear mapping sub-network in ContextNet block. We conduct extensive experiments on four real-world datasets and the experiment results demonstrate that our proposed ContextNet-PFFN and ContextNet-SFFN model outperform state-of-the-art models such as DeepFM and xDeepFM significantly., Comment: arXiv admin note: text overlap with arXiv:2102.07619

Published
2021

7. Markowitz-Based Portfolio Research

Author

Lin, Yuyang, Su, Guoxi, Pan, Jingrui, Zhang, Pengtao, Dou, Runliang, Editor-in-Chief, Liu, Jing, Editor-in-Chief, Khasawneh, Mohammad T., Editor-in-Chief, Balas, Valentina Emilia, Series Editor, Bhowmik, Debashish, Series Editor, Khan, Khalil, Series Editor, Masehian, Ellips, Series Editor, Mohammadi-Ivatloo, Behnam, Series Editor, Nayyar, Anand, Series Editor, Pamucar, Dragan, Series Editor, Shu, Dewu, Series Editor, Hemachandran, K., editor, Boddu, Raja Sarath Kumar, editor, and Alhasan, Waseem, editor

Published
2023
Full Text
View/download PDF

9. Correct Normalization Matters: Understanding the Effect of Normalization On Deep Neural Network Models For Click-Through Rate Prediction

Author

Wang, Zhiqiang, She, Qingyun, Zhang, PengTao, and Zhang, Junlin

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

Normalization has become one of the most fundamental components in many deep neural networks for machine learning tasks while deep neural network has also been widely used in CTR estimation field. Among most of the proposed deep neural network models, few model utilize normalization approaches. Though some works such as Deep & Cross Network (DCN) and Neural Factorization Machine (NFM) use Batch Normalization in MLP part of the structure, there isn't work to thoroughly explore the effect of the normalization on the DNN ranking systems. In this paper, we conduct a systematic study on the effect of widely used normalization schemas by applying the various normalization approaches to both feature embedding and MLP part in DNN model. Extensive experiments are conduct on three real-world datasets and the experiment results demonstrate that the correct normalization significantly enhances model's performance. We also propose a new and effective normalization approaches based on LayerNorm named variance only LayerNorm(VO-LN) in this work. A normalization enhanced DNN model named NormDNN is also proposed based on the above-mentioned observation. As for the reason why normalization works for DNN models in CTR estimation, we find that the variance of normalization plays the main role and give an explanation in this work.

Published
2020

22. The complete chloroplast genome sequence of Pinus bhutanica (Pinaceae) and its phylogenetic implications.

Author

Lv, Shiqi, Chen, Jiao, Li, Bingbing, Fu, Taotao, Song, Mingliang, Zhang, Pengtao, Liu, Kang, Kou, Yixuan, and Wang, Jing

Subjects
WHOLE genome sequencing, CHLOROPLAST DNA, PINE, PINACEAE, WHITE pine, TRANSFER RNA
Abstract

Pinus bhutanica is a critically endangered conifer and occurs only in central Bhutan, northwestern Yunnan and southeastern Xizang in China. In this study, the complete chloroplast genome of Pinus bhutanica was first assembled based on next-generation sequencing. The genome sequence was 116,919 bp in length with an overall GC content of 38.75%. A total of 106 functional genes were detected in the genome, including 72 protein-coding genes (PCGs), 30 transfer RNA (tRNA) genes, and four ribosomal RNA (rRNA) genes. The phylogenetic tree reconstructed by 12 chloroplast genomes revealed that P. bhutanica is most closely related to Pinus wangii in subsection Strobus of Pinus. [ABSTRACT FROM AUTHOR]

Published
2024
Full Text
View/download PDF

27. Agricultural Land Quality Evaluation and Utilization Zoning Based on the Production–Ecology–Health Dimension: A Case Study of Huanghua City.

Author

Wang, Fan, Zhang, Pengtao, Zhang, Guijun, and Cui, Jiahao

Subjects
FARMS, EVALUATION utilization, LAND use, ALKALINE hydrolysis, GROUNDWATER quality
Abstract

Clarifying the constituent elements of agricultural land quality, carrying out multi-dimensional quality evaluation of agricultural land, and implementing precise land consolidation and utilization zoning all have important guiding significance for achieving efficient utilization of agricultural land in China. This work analyzed the multi-dimensional evaluation framework of agricultural land based on its comprehensive quality elements and the production, ecological, and health functions. This paper constructed a multi-dimensional agricultural land "production–ecology–health" quality evaluation index system and evaluation criteria, and carried out a multi-dimensional quality evaluation of agricultural land in Huanghua City, Hebei Province, China. The spatial superposition of each dimension′s evaluation results, combined with the logical relationship between agricultural land use and each dimension′s quality, realized the renovation and utilization zoning of agricultural land. The results are as follows: (1) The production, ecological, and health qualities of agricultural land in Huanghua City were below the average and there is spatial variability, whose proportions of grade III and below were 63.12, 66.23, and 69.32%, respectively. In addition, low score areas are mainly located in the south and northwest of the study area. (2) The obstacle factors to quality in different dimensions were different: The obstacle factors to production quality were matter content, soil pH, irrigation guarantee rate, and alkaline hydrolysis nitrogen; groundwater salinity and depth, soil pH, and chemical fertilizers consumption for ecological quality; and groundwater salinity and depth and soil pH for health quality. (3) Agricultural land in Huanghua city is divided into five types of remediation, including 30,277.34 hm2 for high efficiency utilization area, 10,576.54 hm2 for production quality cultivation area, 34,387.86 hm2 for health quality cultivation area, and 56,311.22 hm2 for comprehensive consolidation and restoration area; special remediation measures are proposed for different types of zones. The work improves the multi-objective quality evaluation index system for agricultural land and implements differentiated land remediation strategies by identifying obstacle factors through zoning. It provides methodological ideas to improve the efficiency of land remediation and utilization. [ABSTRACT FROM AUTHOR]

Published
2023
Full Text
View/download PDF

28. Experimental study and mechanism analysis on basic mechanical properties of basalt fiber reinforced concrete.

Author

Wei, Chengpeng, Sun, Xinjian, Yu, Zhenpeng, and Zhang, Pengtao

Subjects
FIBER-reinforced concrete, BASALT, STRESS-strain curves, TENSILE strength, FAILURE mode & effects analysis
Abstract

In order to comprehensively examine the influence of basalt fiber on the basic mechanical properties of concrete, the axial compression, axial tension and direct shear tests were carried out on basalt fiber reinforced concrete (BFRC) with different fiber contents and fiber lengths. The failure modes and stress–strain curves under different loading modes were obtained from the test data, and the basic mechanical parameters of BFRC were extracted from the curves. Based on the results, the following conclusions were drawn. With the increase of fiber content, both the compressive strength and tensile strength showed an upward trend first followed by a downward trend. In general, the shear strength of BFRC was lower than that of ordinary reinforced concrete with no fiber (RC). All BFRC specimens had an increased peak strain and significantly improved deformability compared to RC. Based on a comprehensive analysis, the specimen with a fiber content of 0.2% and a fiber length of 6 mm (BFRC‐0.2%‐6) exhibited the best mechanical properties. In addition, mathematical regression analysis was performed on the normalized stress–strain curves of BFRC, and the influencing parameters model of basalt fiber was proposed based on the fiber content and fiber length. Moreover, scanning electron microscopy (SEM) and computed tomography were utilized to analyze the influencing mechanism of basalt fiber content and fiber length on the mechanical properties of concrete from the microcosmic and mesoscopic perspectives. [ABSTRACT FROM AUTHOR]

Published
2023
Full Text
View/download PDF

30. FiBiNet++:Improving FiBiNet by Greatly Reducing Model Size for CTR Prediction

Author

Zhang, Pengtao and Zhang, Junlin

Subjects
FOS: Computer and information sciences, Artificial Intelligence (cs.AI), Computer Science - Artificial Intelligence, Information Retrieval (cs.IR), Computer Science - Information Retrieval
Abstract

Click-Through Rate(CTR) estimation has become one of the most fundamental tasks in many real-world applications and various deep models have been proposed to resolve this problem. Some research has proved that FiBiNet is one of the best performance models and outperforms all other models on Avazu dataset.However, the large model size of FiBiNet hinders its wider applications.In this paper, we propose a novel FiBiNet++ model to redesign FiBiNet's model structure ,which greatly reducess model size while further improves its performance.Extensive experiments on three public datasets show that FiBiNet++ effectively reduces non-embedding model parameters of FiBiNet by 12x to 16x on three datasets and has comparable model size with DNN model which is the smallest one among deep CTR models.On the other hand, FiBiNet++ leads to significant performance improvements compared to state-of-the-art CTR methods,including FiBiNet.

Published
2022

31. A Danger Feature Based Negative Selection Algorithm

Author

Zhang, Pengtao, Tan, Ying, Hutchison, David, editor, Kanade, Takeo, editor, Kittler, Josef, editor, Kleinberg, Jon M., editor, Mattern, Friedemann, editor, Mitchell, John C., editor, Naor, Moni, editor, Nierstrasz, Oscar, editor, Pandu Rangan, C., editor, Steffen, Bernhard, editor, Sudan, Madhu, editor, Terzopoulos, Demetri, editor, Tygar, Doug, editor, Vardi, Moshe Y., editor, Weikum, Gerhard, editor, Tan, Ying, editor, Shi, Yuhui, editor, and Ji, Zhen, editor

Published
2012
Full Text
View/download PDF

33. An Immune Concentration Based Virus Detection Approach Using Particle Swarm Optimization

Author

Wang, Wei, Zhang, Pengtao, Tan, Ying, Hutchison, David, editor, Kanade, Takeo, editor, Kittler, Josef, editor, Kleinberg, Jon M., editor, Mattern, Friedemann, editor, Mitchell, John C., editor, Naor, Moni, editor, Nierstrasz, Oscar, editor, Pandu Rangan, C., editor, Steffen, Bernhard, editor, Sudan, Madhu, editor, Terzopoulos, Demetri, editor, Tygar, Doug, editor, Vardi, Moshe Y., editor, Weikum, Gerhard, editor, Tan, Ying, editor, Shi, Yuhui, editor, and Tan, Kay Chen, editor

Published
2010
Full Text
View/download PDF

35. MemoNet:Memorizing Representations of All Cross Features Efficiently via Multi-Hash Codebook Network for CTR Prediction

Author

Zhang, Pengtao and Zhang, Junlin

Subjects
FOS: Computer and information sciences, Computer Science - Machine Learning, Artificial Intelligence (cs.AI), Computer Science - Artificial Intelligence, Information Retrieval (cs.IR), Computer Science - Information Retrieval, Machine Learning (cs.LG)
Abstract

New findings in natural language processing(NLP) demonstrate that the strong memorization capability contributes a lot to the success of large language models.This inspires us to explicitly bring an independent memory mechanism into CTR ranking model to learn and memorize all cross features'representations. In this paper,we propose multi-Hash Codebook NETwork(HCNet) as the memory mechanism for efficiently learning and memorizing representations of all cross features in CTR tasks.HCNet uses multi-hash codebook as the main memory place and the whole memory procedure consists of three phases: multi-hash addressing,memory restoring and feature shrinking.HCNet can be regarded as a general module and can be incorporated into any current deep CTR model.We also propose a new CTR model named MemoNet which combines HCNet with a DNN backbone.Extensive experimental results on three public datasets show that MemoNet reaches superior performance over state-of-the-art approaches and validate the effectiveness of HCNet as a strong memory module.Besides, MemoNet shows the prominent feature of big models in NLP,which means we can enlarge the size of codebook in HCNet to sustainably obtain performance gains.Our work demonstrates the importance and feasibility of learning and memorizing representations of all cross features ,which sheds light on a new promising research direction.

Published
2022
Full Text
View/download PDF

44. Noncovalent microarrays from synthetic amino-terminating glycans: Implications in expanding glycan microarray diversity and platform comparison

Author

Li, Chunxia, primary, Palma, Angelina S, additional, Zhang, Pengtao, additional, Zhang, Yibing, additional, Gao, Chao, additional, Silva, Lisete M, additional, Li, Zhen, additional, Trovão, Filipa, additional, Weishaupt, Markus, additional, Seeberger, Peter H, additional, Likhosherstov, Leonid M, additional, Piskarev, Vladimir, additional, Yu, Jin, additional, Westerlind, Ulrika, additional, and Chai, Wengang, additional

Published
2021
Full Text
View/download PDF

47. Attack-defense game for critical infrastructure considering the cascade effect

Author

Wu Tao, Sun Yun, Zhong Jilong, Gao Yangjun, Fu Chaoqi, and Zhang Pengtao

Subjects
Mathematical optimization, Strategy, Optimization problem, Computer science, Stochastic game, Offensive, Resource allocation, Safety, Risk, Reliability and Quality, Game theory, Industrial and Manufacturing Engineering, Critical infrastructure, Cascade effect
Abstract

The important status of critical infrastructure makes it a target of attacks in the new era. Based on game theory, we analyze the offensive and defensive issues of critical infrastructure from a network perspective, and we propose a network offensive and defensive game model considering the impact of the cascade effect. We first establish a two-player game model for an attacker and a defender acting simultaneously. The nodes are categorized to achieve the transformation of the focus of the game from decision-making for the nodes to resource allocation, thereby simplifying the calculation space of the network game strategy. Both pure strategy and mixed strategy are carried out for the theoretical analysis and simulation of four kinds of optimization problems. With the pure strategy, the initial decision equilibrium is that the attacker's strategy is to attack the Equal Cost-Effective ratio nodes (ECE nodes), and the defender defends the High Cost-Efficient ratio nodes (HCE nodes). At this point, the defender's payoff is 0. With the mixed strategy, different optimization problems have different equilibrium solutions. In particular, the attacker adds a new equilibrium solution for the strategy with the cascading effect, which is related to the structural properties of the network. For the defender, the strategy is different when maximizing his/her own profit or minimizing the attacker's benefit. This can explain how the attacker and the defender can achieve a win-win situation.

Published
2021

48. Camouflage strategy of a Stackelberg game based on evolution rules

Author

Fu Chaoqi, Zhang Pengtao, Gao Yangjun, Zhou Lin, and Du Na

Subjects
Computer science, General Mathematics, Applied Mathematics, Node (networking), Offensive, General Physics and Astronomy, Statistical and Nonlinear Physics, Penalty coefficient, Computer security, computer.software_genre, Critical infrastructure, Cascading failure, Camouflage, Optimal allocation, Stackelberg competition, computer
Abstract

The importance of critical infrastructure makes it a target of attacks in the new era. Using the Stackelberg game model, we analyze the offensive and defensive security issues of critical infrastructure from a network perspective. Considering the impact of cascading failure, nodes are divided into two categories according to the cost-efficiency ratio. High cost-efficiency (HCE) nodes are the initial protection nodes, and each initial protection node sets up an agent to manage the defense resources. The defense resources can achieve two effects: protection and camouflage. Complete defense can avoid failure after being attacked. Partial defenses can avoid the attacker's attack by camouflage, and the effect of camouflage is related to the amount of defense resources. We propose an evolution model to realize the optimal allocation of resources through the evolutionary game of the agents, and analyze the influence of three parameters on evolution. The results show that the initial willingness of the agent will affect the result of evolution only when it is small. If the willingness exceeds a critical value, the result of evolution is only affected by the penalty coefficient and camouflage coefficient. To optimize the overall defense effect, the penalty coefficient and camouflage coefficient must cooperate, which is explained by the causal relationship between the agent's behavior and decisions.

Published
2021

49. Simulation of Spatial Service Range and Value of Carbon Sink Based on Intelligent Urban Ecosystem Management System and Net Present Value Models—An Example from the Qinling Mountains.

Author

Ma, Xinping, Li, Jing, Zhao, Kuifeng, Wu, Tao, and Zhang, Pengtao

Subjects
ECOSYSTEM management, ECOSYSTEM services, NET present value, CARBON cycle, ENERGY consumption of lighting, CARBON offsetting, DIFFUSION coefficients, CARBON sequestration
Abstract

Under the current background of carbon neutrality, the quantification of carbon neutrality in Qinling Mountains, the calculation of the spatial service scope, and the estimation of economic value are of great significance for the improvement of carbon neutrality spatial planning and the carbon trading market. The Intelligent Urban Ecosystem Management System (IUEMS) was used to calculate the carbon sequestration of the Qinling Mountains. The carbon emission of Qinling Mountains was spatialized through model fitting, using the night light data and energy consumption data of various cities. The static carbon content of the Qinling Mountains was obtained. Combined with the common gas diffusion coefficient, under normal temperature and pressure, the service range of the Qinling Mountains' carbon neutralization to the surrounding area was calculated. The spatial distribution of carbon sequestration in 2030 and 2050 was simulated using Patch generating Land Use Simulation (PLUS) model and Net Present Value (NPV) model. Under two discount rates, the model could calculate the carbon value in 2012, 2016, 2020, 2030, and 2050, respectively. The results proved that the value of carbon in the Qinling Mountains is gradually increasing. Since 2012, the service range of carbon neutralization in the Qinling Mountains was 175–262.63 km, and the service range has been gradually expanded. The discount rate is inversely proportional to the carbon value. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

50. Piezomagnetic Anomalies Associated with the 2021 M W 7.3 Maduo (China) Earthquake.

Author

Song, Chengke, Zhang, Pengtao, Wang, Can, and Chu, Fei

Subjects
GEOMAGNETISM, EARTHQUAKES, MAGNETIZATION
Abstract

Stress changes due to earthquake rupture can disturb geomagnetic fields significantly. In order to investigate the impact of the 2021 MW 7.3 Maduo earthquake on geomagnetic fields, a piezomagnetic model is constructed based on the coseismic slip to calculate the static coseismic piezomagnetic anomalies (PMs). The PMs are considerable in near-field. However, the PMs are negligible in regions tens of kilometers from the fault rupture. The PMs of our model are consistent with those of other strike-slip earthquakes, indicating that our piezomagnetic model is reasonable. The east component of observed coseismic geomagnetic changes and calculated PMs on a geomagnetic repeat station located about 6 km from fault trace are +4.8 ± 2.2 nanotesla and +4.3 nanotesla, respectively. It seems that the piezomagnetic model can explain the observed data. The PMs are up to 10 nanotesla in the near-field with the initial magnetization of 3 A/m and stress sensitivity of 2 × 10−3 MPa−1. Consequently, considerable coseismic geomagnetic changes that are above error could be observed along the fault, especially at locations with geometrical complexities. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results