Заседание научного семинара лаборатории ЛАТАС

Models such as residual networks, recurrent neural network decod-ers, and normalizing flows build complicated transformations by composing a sequence of transformations to a hidden state: , where and .These iterative updates can be seen as an Euler discretization of a continuous transformation. What happens as we add more layers and take smaller steps? This idea is a foundation of Neu-ral Ordinary Differential Equation (NeuralODE), which parameterized the con-tinuous dynamics of hidden units using an ordinary differential equation speci-fied by a neural network .

NeuralODE is a powerful tool for dealing with real physical processes, where data is naturally consistent and continuous, in the sense that the data points are observations of a continuity changing state. As a demonstration of the power of the NeuralODE method we will consider a problem of establishing parameters of atmospheric thermal radiation.

Приглашаются все желающие!
Явка аспирантов школы компьютерных наук обязательна.

Для участия, пожалуйста, зарегистрируйтесь