In recent years, machine learning (ML) algorithms have evolved at a very fast pace, revolutionizing, along the way, numerous sectors of modern society. ML has found countless applications in our daily lives, making it almost impossible to describe all of its uses. Notably, artificial neural networks (NNs) stand out as one of the most powerful and diverse classes of models. The NN-empowered tools assist in navigating our routes to the target destinations, providing personalized recommendations for entertainment, suggesting shopping preferences, classifying emails, translating text, and can even mimic human interactions in the form of chat bots. All of these applications are inspired by the same idea: using artificial intelligence can enhance our lives and boost efficiency when dealing with these tasks. The scientific community has seen a boom in machine learning studies, and many of the latest NN-based models outperform the traditional approaches by a very large margin. Therefore, the potential of integrating NN models into various scientific applications is boundless.

At the same time, NNs are usually criticized for being “black-box” models that are hard to interpret and understand, with an aura of mystery surrounding these algorithms. In this short course, we will delve into the foundations of neural networks, emphasizing approaches and best practices to model training, independent validation and testing, as well as model deployment. We will describe both the basic concepts and building blocks of the neural network architectures, and also touch upon the more advanced models. Our objective is to explain how neural network models can be understood in comprehensive but relatable terms for participants coming from a broad range of backgrounds.