SORN-based Cascade Support Vector Machine
Nils Hülsmeier, Moritz Bärthel, Jochen Rust, Steffen Paul · 2020
This paper presents a SORN-based cascade support vector machine (SVM). For the first time a SVM is implemented by the usage of SORNs (Set Of Real Numbers). The SORN representation is a dual number representation that uses a fixed set of exact values and open intervals. Arithmetic operations in SORN representation can be realized by lookup tables which allows fast and low-complexity computing. This arithmetic is used for the non-support vector filtering in the early stages of a cascade SVM. Once the training subsets passed the bottom layer of the cascade, the optimization on the remaining support vectors can be done in classic representations like floating point or fixed point.