Yesterday, September 28, Library on Gazetny held a lecture "Basics of neural networks." Vladimir Kosarev, Researcher of the Industrial Organization and Infrastructure Economics Department at the Gaidar Institute, told the audience about how neural networks evolved: from the simplest mathematical models to modern large language models.
The lecture participants traced the history of technology development from the McCulloch-Pitts formal neuron to modern LLMs. The lecturer explained how approaches to training have changed and why the increase in computing power and data volumes has become a critical factor for the success of modern neural networks.
The central part of the lecture was a practical introduction to how a neural device works. Using a simplified task of controlling a self-driving car, Vladimir Kosarev showed how the network uses data, makes predictions, and adjusts its parameters to minimize errors.
The expert also talked about image classifications and explained how a computer learns to distinguish handwritten digits and recognize objects in photos. Besides that, they touched on the use of such models in practical economics and finance tasks.
The lecture participants showed high engagement: attendees asked questions about the prospects for the development of neural networks, concerns related to artificial intelligence, as well as its application in healthcare and culture.