How a 1990s textbook explains the structure of modern neural networks

The author shares an interesting observation made while studying an old 1990s linear algebra textbook. The text provides an example of representing chemical molecules as vectors, which seems unusual at first glance. However, a deeper analysis shows that this approach essentially anticipated modern machine learning methods, specifically the concept of representation learning. The author concludes that many principles underlying modern neural networks have deep roots in classical linear algebra. The article emphasizes that the ability to translate any subject area into the language of vectors is a fundamental basis for artificial intelligence algorithms. This historical excursion helps to better understand that modern technologies are not magic, but a logical development of mathematical methods that have been used for decades, simply under different names.