神经网络处理非线性动态信号的最新进展

需积分: 4 125 浏览量更新于2024-07-18 收藏 9.34MB PDF 举报

《神经进阶：处理非线性动态信号》是SmartInnovation、Systems and Technologies系列的第102卷，由Anna Esposito、Marcos Faundez-Zanuy、Francesco Carlo Morabito和Eros Pasero共同编辑。该电子书聚焦于智能系统领域，特别关注非线性动态信号处理的最新研究进展。系列主编罗伯特·詹姆斯·霍尔韦特（Bournemouth University和KES International）以及拉克希米·C·贾恩（University of Technology Sydney、University of Canberra和KES International）共同负责，旨在通过提供一个平台，汇集跨学科的研究成果，将这些主题的前沿知识以易于理解的方式呈现给公众。本书系列的目标是促进对知识、智能和可持续性的探讨，鼓励多学科交叉研究。《神经进阶：处理非线性动态信号》正是这种跨领域研究的典范，它探讨了如何利用神经网络和其他先进算法来解析和分析复杂且变化无常的动态数据，例如生物电信号、环境监测数据或工业设备的运行数据。非线性动态系统的特点在于它们的行为不能简单地用线性模型描述，而是受多种因素交互影响，这使得处理这类信号成为许多现代技术领域（如机器学习、控制工程、信号处理和人工智能）的重要挑战。书中可能涉及的内容包括但不限于：非线性动力学的基本理论，如混沌理论和分形；深度学习在非平稳信号识别中的应用，如卷积神经网络和递归神经网络；时间序列分析方法，如短时傅立叶变换或小波分析；以及适应性和自适应控制策略，以应对非线性系统的不确定性。此外，可能还会讨论实时信号处理技术，如在线学习和流处理，以便实时响应和优化动态环境中的系统性能。对于研究人员和工程师来说，这本书提供了宝贵的参考资料，帮助他们理解如何利用神经计算方法提升在航空航天、能源、通信、医疗和自动化等领域的信号处理能力，推动科技进步和社会可持续发展。《神经进阶：处理非线性动态信号》不仅是一本深入的学术著作，还是一份实用的工具指南，旨在启发创新思维，解决实际问题，并促进非线性信号处理领域的进一步发展。

4 A. Esposito et al.

of industrial and economic applications. The collection of chapters presented in this

book offer a scenery of the current progresses in such scientiﬁc domain.

Keywords Financial and industrial process

· Speech enhancement

Machine leaning methods

· Neural networks · Artiﬁcial intelligence

Nonlinear signal processing

· Decision making in economic and ﬁnance

Evolutionary algorithms

1.1 Introduction

Research on Computational Intelligence (CI) and Information Communication Tech-

nologies (ICT) aims to improve the quality of life of the end users by implementing

user friendly complex autonomous systems that simplify the solution of problems

related to their everyday needs. The difﬁculties associated with the dealing of non-

linear data i s very challenging in these ﬁelds for practical and intellectual reasons.

From the practical point of view, solving such hindrances will improve produc-

tivity and facilitate the user interaction with engineering systems.

Intellectually, to accurately process nonlinear data hold considerable promises as

well as challenges, in the years to come, for scientists and developers alike. Consid-

ering the immense amount of research over the last three decades, one may wonder

why the processing of nonlinear data is still an unsolved problem.

This is because, when dealing with different nonlinear data, from an engineering

design point of view, there is the need to deﬁne a model for each data category.

When the number of data increase, the computational effort to process and handle

relations among them, increase exponentially. Basic research may continue to provide

mathematical models for distinct nonlinear categories. However, the exponential

complexity required by large amount of data with sophisticated relations among

them cannot be easily solved both in the short and long time period. The challenges

are to handle complex relations among the data and ensure optimality in the system

design.

Since automatize t he nonlinear processing of large amount of data cannot be solved

only specifying data categories and relationships among them, the new approach is

to observe how living creatures successfully interact among them and within an

unpredictable environment. The principles governing such interactions, on the basis

of the daily experience, seem to be that of building iteratively changing models

of the data at the hand, and stores them in terms of iteratively extracted changing

features, and iteratively derived features representations. Then, when required from

the circumstances, choose the model and associated representation that best ﬁt the

task at the hand and assess its likelihood of success on the basis of the limited resource

and available information.

Neural, adaptive systems seem to be the solution more appropriate under such

circumstances. These systems are able to solve the challenge to process nonlinear

data since they show characteristics of adaptation to the environment, distributed pro-

1 Processing Nonlinearities 5

cessing abilities due to their large number of computational elements, and resources,

in term of algorithms and mathematical models, to provide optimal solutions for the

constantly changing data. They allow continuously data learning, and when correctly

applied, provide solutions that outperform sophisticated conventional linear systems.

The revolutionary distinctiveness of such systems with respect to the traditional engi-

neering design approach is in discarding a priori speciﬁcations of their parameters

and exploit external data to ﬁx them automatically. To do so they apply learning

algorithms with feedback loops and cost functions that will converge to the desired

output through a continuously training that modify the system’s parameters exploit-

ing information on unremittingly changing external data. What remain to the designer

is to specify the architecture of the system (essentially the neural network topology),

deﬁne the most effective cost function, and decide the extent to which admit an error

from the system. Even these decisions are not left to the designer alone, since an

extensive research has been conducted on evaluating learning algorithms, training

procedures, and allowed errors for several neural network models and several appli-

cation contexts, such as signal recognition, separations, and enhancement, system

predictions, noise cancellation, patterns’ classiﬁcation, and control [1–4, 6].

1.2 Content of This Book

The research reported in this book bring up the latest advances on neural processing

of nonlinear dynamic signals. The content of the book is organized in sections,

each dedicated to a speciﬁc topic, including peer-reviewed chapters, not published

elsewhere. The inspiring content of each chapters was discussed for the ﬁrst time

at the International Workshop on Neural Networks (WIRN 2017) held in Vietri

sul Mare, Italy, from the 14th to 16th of June 2017. The workshop, is nowadays a

historical and traditional scientiﬁc event gathering together researchers from Europe

and overseas.

Part I describes the main motivations for a book on neural advances in processing

nonlinear dynamic signals through an introductory chapter proposed by Anna Espos-

ito, Marcos Faundez-Zanuy, Francesco Carlo Morabito and Eros Pasero, which are

the editors of the book.

Part II is dedicated to nonlinear data and techniques for data mining. This is

because the need for mining huge and complex data has become essential in educa-

tion, web mining, social network analysis, s ecurity, medicine, and health manage-

ment, as well as in many other engineering ﬁelds. Neural adaptive systems seems

to fail to handle these large amount of data, because they are too big to a certain

extent and hardly scalable. Solutions are however proposed in this section, always

exploiting neural systems and machine learning algorithms which succeed to scale

and mine such data.

Part III is dedicated to computational intelligence and related techniques in indus-

trial and ICT engineering. This section proposes neural adaptive systems able to

manage in a quick, effective and efﬁcient manner the increasing amount of infor-

6 A. Esposito et al.

mation to be shared and processed by organizations and industry, since previously

offered solutions were not able to handle the transformation of large amounts of data

(involving uncertainties) into knowledge. In this context, computational intelligence

techniques are proposed, mainly including artiﬁcial neural network, learning systems,

evolutionary computation, and fuzzy logic, to exactly extract (hidden) information

from data and make it useful for application such as fault diagnosis and forecasting,

product quality and process forecasting, smart maintenance systems, big data and

analytics transforms, knowledge management, production, planning and scheduling,

evolutionary big data interaction for predictive product design and marketing.

Part IV is intended to report on neural intelligent tools for decision making in eco-

nomics and ﬁnance, and the current exploitation of intelligent systems for modeling

economic behaviors which involve either evolutionary optimizers showing collective

intelligence—such as artiﬁcial bee colonies, genetic algorithms, particle swarm opti-

mizations—or machine learning techniques based on supervised and reinforcement

learning procedures, or unconventional fuzzy logic and rough sets. The contribu-

tions of this section embrace these research ﬁelds and highlight the beneﬁts of using

intelligent tools in the investigation of economic/ﬁnancial phenomena.

1.3 Conclusion

When Mc Carthy and colleagues [5] suggested the basic idea of providing a machine

of “intelligence” he may have not been thinking to the processing of nonlinear data

and the detection, recognition, and analysis of complex data structures. He may

have even not been thinking to neural adaptive systems for processing nonlinear

dynamics even though his statement was “that every aspect of learning or any other

feature of intelligence can in principle be so precisely described that a machine

can be made to simulate it”. Currently, neural adaptive processing techniques, have

capacities and skills, such as—reacting ﬂexibly to situations, taking advantage of

fortuitous circumstances, recognizing the relative importance of different elements

in a situation, ﬁnding similarities in different situations, synthesizing new concepts,

taking old concepts and linking them in a different way—which although are not

sufﬁcient to deﬁne “intelligence” wisely, certainly have revolutionized the design and

modeling of complex engineering problems, offering better and more “intelligent”

solutions to them.

Acknowledgements The research leading to the results presented in this paper has been conducted

in the project EMPATHIC (Grant N: 769872) that received funding from the European Union’s

Horizon 2020 research and innovation programme.

剩余312页未读，继续阅读

J-10

粉丝: 18
资源: 483

神经网络处理非线性动态信号的最新进展

Neural Network Methods in Natural Language Processing epub

batch processing in a neural network processor

RBF-neural-network-nonlinear-system.zip_nonlinear control_output

Adaptive neural tracking control of nonlinear time-delay systems with disturbances

Handbook of Neural Network Signal Processing.pdf

Advances in Neural Networks Computational and Theoretical Issues

Advances in Neural Computation, Machine Learning, and Cognitive Research

Nonlinear System Identification using RBF Neural Network:Nonlinear System Identification using RBF Neural Network-matlab开发

思维导图_综述-Recent Advances in Neural Question Generation_.pdf

Adaptive Neural Control for Output Feedback Nonlinear

最新资源