Data Analytics and Management in Data Intensive Domains: 21st International Conference, DAMDID/RCDL 2019, Kazan, Russia, October 15–18, 2019, Revised Selected Papers

Recently, deep neural networks have showed amazing results in many fields. To build such networks, we usually use layers from a relatively small dictionary of available modules (fully-connected, convolutional, recurrent, etc.). Being restricted with this set of modules complicates transferring technology to new tasks. On the other hand, many important applications already have a long history and successful algorithmic solutions. Is it possible to use existing methods to construct better networks? In this paper, we cover three approaches to combining neural networks with algorithms and discuss their pros and cons. Specifically, we will discuss three approaches: structured pooling, unrolling of algorithm iterations into network layers and explicit differentiation of the output w.r.t. the input.