Ekaterina Sirazitdinova (
AI is everywhere! It defines news feeds, recommends products to buy on e-commerce platforms, translates texts into different languages, steers autonomous cars and drones, helps doctors identify diseases and accomplishes a wide range of other tasks. You might have heard that to build a successful AI product, one needs to train an AI model that should be highly accurate, robust to false positives, and have high throughput. But what does it mean to train an AI model? In this talk, we will demystify basic concepts behind AI model training: from the definition of a deep neural network to critical parameters controlling training. We will examine problems typically encountered when training a model and share best mitigation practices.