ABSTRACT
The theoretical foundations of Artificial Neural Networks (ANNs) were established in the 20th century, beginning with the McCulloch–Pitts perceptron, progressing through the development of the multilayer perceptron, and culminating in modern architectures where the backpropagation algorithm proved effective for training. However, this technology has only recently experienced a surge in popularity, driven by the advent of sufficiently powerful Graphics Processing Units (GPUs) capable of training NNs with vast numbers of trainable parameters. Large Language Models (LLMs) are now capable of assisting in numerous human activities, and the demand for other AI-powered systems continues to grow. Data analysis and neural network-based solutions now perform as well as — or often better than — classical algorithms in nearly every domain. As a result, the early 21st century can be considered the rise of the data-driven algorithms paradigm.
