Oliver Gäfvert: An algebraic perspective on artificial neural networks
Time: Fri 2018-04-13 13.15 - 14.15
Location: Room 3418, Lindstedtsvägen 25, 4th floor, Department of Mathematics, KTH
Participating: Oliver Gäfvert
Abstract: I will talk about the algebraic objects associated to polynomial neural networks in the context of classification problems. These include the semi-algebraic set of network parameters solving the classification problem, and the resulting decision boundary of the solution. The focus of the talk is to relate properties of these objects to the architecture of the network, and comparing shallow and deep networks.