Composite Artificial Intelligence
Description
Unlock the full potential of context-aware AI while navigating critical hurdles like bias mitigation and ethical governance with this definitive resource on the future of composite artificial intelligence.
In the rapidly evolving landscape of artificial intelligence, the demand for more adaptive, intelligent, and context-aware systems has led to the emergence of composite artificial intelligence: a paradigm that integrates multiple AI techniques to solve complex real-world problems with higher efficiency and intelligence. This book is a groundbreaking exploration of the next evolution in AI, where diverse methodologies like machine learning, symbolic reasoning, and cognitive computing converge to solve complex, real-world problems with unprecedented intelligence and adaptability. Unlike traditional AI approaches that rely on singular techniques, composite AI harnesses the strengths of multiple paradigms, enabling systems that are more robust, interpretable, and capable of human-like decision-making. This book provides a comprehensive roadmap for understanding and implementing these advanced systems, from foundational theories to cutting-edge applications across industries such as healthcare, finance, and smart manufacturing. It delves into critical challenges, including bias mitigation, integration hurdles, and ethical governance, while showcasing real-world case studies that demonstrate the transformative potential of composite AI. With its balanced blend of theory, technical depth, and actionable insights, this book is a definitive resource for unlocking the full potential of AI in an increasingly complex world.
Readers will find the volume:
- Explores the intersection of machine learning, symbolic reasoning, and cognitive computing for solving real-world challenges smarter and faster;
- Introduces cutting-edge techniques for bias reduction, optimization, and seamless multi-method integration;
- Provides real-world case studies and scalable frameworks to demonstrate how composite AI is transforming industries;
- Presents ethical implications and current innovations to build trustworthy, compliant AI systems that align with regulations;
Audience
Academics, policymakers, AI researchers, data scientists, AI and machine learning engineers and developers, and??industry professionals working in healthcare, finance, manufacturing, and cybersecurity who need robust, explainable, and adaptive AI solutions.
More details
Other editions
Additional editions
Persons
T. S. Arun Samuel, PhD is a Professor on the Department of Electronics and Communication Engineering, National Engineering College, Tamil Nadu, India with more than two decades of experience. He has authored more than 65 research articles published in prestigious international journals, 15 articles in international conferences, one patent, and three edited books. His research interests are focused on advancing the frontiers of nanoelectronic device technologies through innovative modeling and simulation techniques.
L. Jerart Julus, PhD is an Assistant Professor in the Department of IT, School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, India. He is a member of IEEE, the Computer Society of India, and the Optical Society of India. His research interests include multicarrier communications systems, radio over fiber, and visible light communication.
P. Kanimozhi, PhD is a Professor of Computer Science and Engineering, IFET College of Engineering, Tamil Nadu, India with more than 19 years of teaching experience. She has published more than 17 research papers in national and international journals. Her current areas of interest include cloud computing security, data mining, and blockchain.
T. Ananth Kumar, PhD is an Associate Professor of Computer Science and Engineering, IFET College of Engineering, Tamil Nadu, India. He has presented papers in national and international conferences and journals, holds patents in various domains, and has edited six books and numerous book chapters. His fields of interest are networks on chips, computer architecture, and application-specific integrated circuit design.
S. Balamurugan, PhD is the Director of Research at iRCS, an Indian Technological Research and Consulting Firm. He has published more than 100 books, 300 papers in international journals and conferences, and 300 patents. With 20 years of research on various cutting-edge technologies, he provides expert guidance in technology forecasting and decision-making for leading companies and startups.
Content
1
Data Fusion Techniques in Composite AI
S. Sowmyayani1*, D. Dhanya2, J. Kavitha3 and R. Roselinkiruba3
1Department of Computer Science, St. Mary's College (Autonomous), Thoothukudi, Tamilnadu, India
2Dept of Artificial Intelligence and Data Science, Mar Ephraem College of Engineering and Technology, Malankara Hills, Elavuvilai, Marthandam, Tamilnadu, India
3Department of CSE, VelTech Rangarajan Dr. Sagunthala, R&D Institute of Science and Technology, Chennai, Tamilnadu, India
Abstract
Nowadays, data is growing equivalent to or more than the population. At the same time, irrelevant and redundant data occupy more than the relevant ones. Considering these accounts, it is studied that data pre-processing needs special attention for any research to be more accurate. Also, it is necessary to group data according to the research. Hence, this chapter explores a few techniques that can be followed while using composite artificial intelligence (AI). A case study is elaborated to emphasize the need for data fusion and research fusion. An important and well-known Human Action Recognition (HAR) is embedded in Content-Based Video Retrieval (CBVR) to fully utilize composite AI's efficiency. It is achieved by combining a few deep learning models to create a HAR-CBVR system and the results (features) are fused in various ways. Experiments are demonstrated with various standard datasets and the results with fusion methods are compared with a few recent methods. The results proved that the inclusion of data fusion techniques works better than other methods.
Keywords: Composite AI, data fusion, deep learning, prediction
1.1 Introduction
Big Data refers to integrating and analyzing vast quantities of complicated heterogeneous data. Applications for big data analytics offer thorough knowledge gleaned from the vast amount of data at hand. Health officials, physicians and patients will all gain more from the new information gleaned from big data analytics. The amount of data generated in numerical biomedical applications is increasing dramatically due to the rapid development of AI and social networks.
AI-powered predictions can enhance many services by identifying risks early, producing fresh perspectives, tracking the calibre of persons or nature and offering more effective options. Some issues in artificial intelligence are getting high throughput, integrating from several databases and using data mining techniques for high-dimensional data. This chapter provides additional information about big data fusion, or merging data from several databases, which helps policymakers make better decisions in their industry.
As individuals become increasingly open to participating in their own healthcare decisions, AI is becoming increasingly common in healthcare applications. Additionally, patients are more inclined to take the initiative to customize their medical care. More research is required because of the personalization of medical care and treatment. The field of mobile health is expanding and includes gadgets like ubiquitous smartphones, low-power body-area wireless networks, and miniature sensors. Composite AI can use predictive modeling to find people at high risk for particular diseases and recognize patterns. Estimating the chance of contracting an illness entails examining risk factors, medical history, genetic data, and even lifestyle information. This makes it possible to implement early diagnosis, proactive interventions, and customized preventative plans that have the potential to enhance health outcomes significantly.
In addition to storing all data in a publicly accessible, decentralized fashion while maintaining anonymity, blockchain offers several services, such as traceability, integrity, security, and non-repudiation. Data from multiple sources and forms, including structured and unstructured data, are integrated using composite artificial intelligence. By taking a comprehensive approach, all pertinent data are used, resulting in a more thorough and precise analysis.
By integrating several analytical methods, composite AI may extract more profound information from data. This improved data utilization makes more useful information for strategic planning and decision-making possible. Composite AI systems frequently use various data sources. Successful implementation depends on efficient data preparation. Preprocessing, cleaning, feature engineering, and integration are all included in this.
This book discusses some simple designs of data fusion in the internet age when the proliferation of data is becoming more widespread due to the enormous use of mobile devices and combining several sources of data into a single format that the composite AI system may utilize. Figure 1.1 shows the structure of composite AI.
The chapter is arranged as follows. The data fusion techniques in composite AI, their challenges, and their benefits are discussed in Section 1.2. The next section elaborates on the fusion technique with a clear application. Section 1.4 emphasizes the proposed algorithm with several experiments and analyses. The last section concludes the chapter with the path to future research.
1.2 Data Fusion Techniques in Composite AI
There are many ways to fuse data. A popular paradigm for classifying data fusion strategies is Dasarathy's classification [1], a well-known authority on data fusion. He grouped fusion techniques according to the relationship between data intake and output, i.e., data fusion, feature fusion, decision fusion, and hybrid fusion.
Figure 1.1 Composite AI.
Figure 1.2 Data-level fusion.
1.2.1 Data Fusion
This approach aims to enhance the eminence, availability, or completeness of the data for further work. Figure 1.2 shows the data-level fusion. Typical techniques for data-level fusion consist of:
- Averaging fusion: This technique takes the mean value of the data from several sources to increase accuracy and lower noise. It works particularly well when sources have random variability and similar properties.
- Concatenation fusion: It combines raw data sequences from various sensors to create a longer sequence. This technique is frequently applied when sensors measure the same phenomenon from various angles. Concatenation is frequently used to enable thorough analysis of data from spectroscopy or image sensors. Several concatenation fusion methods include weighted averaging, Principal Component Analysis, Independent Component Analysis, etc.
- Fuzzy logic fusion: This method handles uncertainty by combining data based on fuzzy logic concepts. It can be helpful when sources offer ambiguous or insufficient information.
1.2.2 Feature Fusion
This approach combines features retrieved from several data sources. This may entail gathering pertinent features such as text, audio, or image from every data source and integrating them into a 1-D feature vector. The goal of this technique is to compile pertinent data from several sources into a form that can be utilized for further work. Figure 1.3 shows the feature-level fusion. Typical techniques for feature-level fusion consist of:
- Features Concatenation: Concatenation combines features from various sources into a 1-D feature vector. When dimensions are comparable or features are the same, it works well.
Figure 1.3 Feature-level fusion.
- Feature selection: The pertinent features from each source are selected according to how crucial they are to the fusion process. By removing unnecessary or uninformative features, it lowers the dimensions. It enhances the fusion process by decreasing noise and increasing efficiency.
- PCA: The original features are turned into orthogonal features called principal components. It can lower the dimension of the data by collecting the greatest amount of variance in the data. It works well at mitigating multicollinearity and lowering the complexity of high-dimensional data.
- Wavelet Transform: By breaking them down into distinct frequency components, the wavelet transform obtains low-frequency and high-frequency intricacies of features. It identifies regional variances and patterns that other approaches might overlook. Applications involving image analysis and signal processing can benefit from it.
1.2.3 Decision Fusion
It is the process of combining judgments or outputs from several datasets to reach a conclusion or inference. Combining decisions from various classifiers, decision-making algorithms, or decision rules is one way to do this. It seeks to enhance the final decision's or inference's accuracy, dependability or robustness by combining the outputs of several datasets. The methods used in this type of fusion include:
- Majority voting: Choosing the option that receives the maximum votes allows decisions from several sources to be combined. It is helpful when sources are occasionally prone to errors and have comparable reliability.
- Weighted Voting: One method that is comparable to majority voting is weighted voting. Decisions are weighted differently in this case, depending on how reliable the source is. It lessens the impact of unreliable sources while increasing the influence of trustworthy ones.
- Plurality voting: It selects the option with the most votes. This method is useful if several classes have no obvious majority for one class.
- Expert opinion aggregate: It combines the opinions...
