This chapter is an analytical report of the way in which data are generated, collected, processed, and used in the consumer device ecosystem. It starts with classifying the different types of devices, which produce varying streams of data, through the use of smartphones, wearables, and smart home systems, among others. It then moves on to discuss the techniques of data collection, which include passive, active, and automated, their technical mechanisms, merits, and consideration to privacy. It also categorizes the information gathered as data relating to behavioral, biometric, location, environmental, and transactional/interaction data, which provide distinctive information to be used by analytics. It also enters into data processing and storage update and storage infrastructures where it highlights its emphasis on edge computing and the cloud, data encryption, and the data lifecycle. The last section shows real-world applications as applied in various sectors, including healthcare, personalized marketing, environmental monitoring, and financial services, based on which the remarkable power of consumer-generated data can be observed. The chapter brings together both technical and ethical insights, to provide the reader with the comprehensive vision of the opportunities and issues related to the use of consumer device data in the context of innovation and decision-making purposes.
S. B. Donaev, A. V. Alizadeh, M. Singh et al.· 0 citations
In this chapter, the author provides an overview of the guiding principles of machine learning (ML), representation, optimization, and generalization, in the well-designed data pipelines. It reviews the bias-variance trade-off as a theoretical model that helps gauge the development of the model and how they perform. It speaks about the family of supervised learning, unsupervised learning, reinforcement learning; complementary and emergent methods; and emergent paradigm in semisupervised and self-supervised emergent learning and hybrid learning classes. The spectrum of algorithms—the linear models, kernel methods, ensembles, probabilistic models, and deep learning structures—is compared on the basis of the inductive biases and the suitability of application. Harshness in model selection, right metrics to use within tasks, and right validation practice are highlighted where you have certain surety of the reliability of the real-world performance. The operational considerations are also the role of MLOps in reproducibility and monitoring and lifecycle. Factors such as equity, accountability, and openness belong to the sphere of ethics and governance that are highlighted as a central part and parcel of ethical ML implementation. Privacy-preserving federal learning: The collaborative modeling technique referred to as federal learning takes advantage of the strengths of privacy-preserving procedures such as federally private training. The strategies that are presented as options to decrease the latency and environmental impact are the efficiency-oriented approaches such as pruning, quantization, and distillation. Streaming inference, edge and cloud codesign, and online learning are applicable on the resource limited and real-time applications. Some of the developments expected are as follows: casual inference, measure of uncertainty, neurosymbolism artificial intelligence, and automatic experimentation. The importance of ensuring that technical rigors become alignable to the individual space situated within context, human value, and governance to produce effective credible outcomes has been emphasized at the close of the chapter.
R.G. Abaszade, A. V. Alizadeh, M. Singh et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
