Generating High-Order Threshold Functions with Multiple Thresholds
This work addresses a specific problem in neural network optimization for researchers in machine learning, but it appears incremental as it builds on existing threshold function methods.
The paper tackles the problem of generating high-order threshold functions with multiple thresholds to optimize high-order neural networks, showing that functions with the same weight but different thresholds can be easily obtained and the network order can be extended while preserving function structure.
In this paper, we consider situations in which a given logical function is realized by a multithreshold threshold function. In such situations, constant functions can be easily obtained from multithreshold threshold functions, and therefore, we can show that it becomes possible to optimize a class of high-order neural networks. We begin by proposing a generating method for threshold functions in which we use a vector that determines the boundary between the linearly separable function and the high-order threshold function. By applying this method to high-order threshold functions, we show that functions with the same weight as, but a different threshold than, a threshold function generated by the generation process can be easily obtained. We also show that the order of the entire network can be extended while maintaining the structure of given functions.