A1056
Title: Theory of structured deep neural networks
Authors: Dingxuan Zhou - University of Sydney (Australia) [presenting]
Abstract: Deep learning based on deep neural networks with network architectures has been powerful in practical applications but is less understood theoretically. The network structures give essential difficulty. An important family of structured deep neural networks is deep convolutional neural networks with convolutional structures. The convolutional architecture is key for computational efficiency but raises scientific challenges. A mathematical theory of approximating and learning functions or operators by deep structured deep neural networks is described.
