Improving classification models with context knowledge and variable activation functions
Andrea Apicella · 2018
This work proposes two methods to boost the performances of a given classifier: the first one, which works on a Neural Network classifier, is a new type of trainable activation function, that is a function which is adjusted during the learning phase, allowing the network to exploit the data better respect to use a classic activation function with fixed-shape; the second one provides two frameworks to use an external knowledge base to improve the classification results.