Neural Ordinary Differential Equations

A paper released in June of 2018 is making headlines as one of the best Artificial Intelligence papers of the year. The work done was by researchers out of the University of Toronto. They developed a new type of neural network (like convolution and recurrent networks) that functions extremely well when performing time-series tasks (which were previously done best by recurrent networks). This new type of network is based on differential equations, which means that things are being integrated within the network. If you are interested, it's a bit of a dense read, but this technology seems very promising. I would suggest reading the two links below in order. The first will provide some important background knowledge that you need to have before reading the second. It should be noted that both links go to PDFs of research papers, so take your time reading through them.

1. Residual Network
2. Neural Ordinary Differential Equations