Lightweight Stochastic Configuration Network Model Driven by Indicator Function Encoding

In order to solve the problem of many parameters and hardware requirements of neural networks in the deployment of edge computing devices. This paper proposes a lightweight stochastic configuration network model driven by indicator function encoding (LSCN-IFE). The model adopts a novel encoding strategy to map the input data into a bit vector through an indicator function, and uses a stochastic configuration algorithm to adaptively determine the interval. Furthermore, hyperdimensional computing is used to optimize the hidden layer node representation, so that network nodes only rely on addition and XOR (Exclusive OR) operations, thereby significantly reducing computational complexity. Theoretical analysis proves that the model has convergence under appropriate conditions. Experimental results show that the model can achieve excellent prediction accuracy, significantly reduce power consumption, and effectively improve real-time prediction capabilities in multiple benchmark data sets and actual industrial applications. This research provides an efficient and scalable solution for neural network deployment in edge computing.