Design and implementation of a customized library for creating versatile neural networks

Read full text Summarise with VIRAT
Authors: Αλέξανδρος Χ. Βασιλείου
Cited by: 0
References: 9

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The modern era, is defined by a technological revolution driven by the emergence of autonomous systems, operated through machine learning and artificial intelligence.A foundational element of these fields is neural networks.A thorough understanding of their operational mechanisms, is critically important for proper utilization and the development of improved technologies that will shape the future.To this end, a customized library has been implemented, aiming to create versatile neural networks, using TypeScript alongside the Node.jsruntime.During development, the mathematical principles underlying neural networks were detailed, including calculus concepts such as linear and non-linear functions, derivatives, partial derivatives, the chain rule, and linear algebra topics like matrices and their properties.The examination focused, on how these mathematical concepts are combined in neural network algorithms, leading to the development of concepts like neurons and network layers, weights, biases, activation and loss functions and the process of training a neural network through backpropagation and gradient descent optimization algorithms.Subsequently, a software architecture plan was created, emphasizing the importance of object-oriented programming, in such a project and the role of each object model forming the library.The development process, highlighted the significance of a deep understanding of the operational mechanisms of neural networks, focusing on making informed decisions, performing preliminary dataset analysis, and understanding the consequences of negligence.Issues, such as exploding and vanishing gradients were identified.Finally, the library package was created, with online distribution capability via npm.

Fields of Study & Subjects

Neural Networks and Applications Artificial Intelligence Applications Machine Learning and ELM

Publication details

Journal / source
University of Thessaly Institutional Repository (University of Thessaly)
Publication date
1 January 2026
Language
EL
Document type
Thesis