Hilbert matrix-based weight initialization enhanced by mutual information for neural network optimization

The wide community of researchers has embraced Artificial Neural Networks (ANNs) to optimize several activities that include approximation alongside regression models. The training efficiency of ANNs depends heavily on the methods used to initialize the weights. The suggested weight initialization system develops the Hilbert matrix method to accelerate training convergence. The implementation of Mutual Information (MI) enables feature selection through the MI score ranking of features. The ordered features are distributed across a scaled Hilbert matrix to assign higher weight to higher-ranked elements and lower weight to lower ones, which results in more rapid training efficiency. This work achieves its main innovation through the combination of Mutual Information-based feature ranking together with Hilbert-matrix-based weight initialization procedures. The combined approach produces an initialization technique that advances convergence speed and strengthens learning stability. The experimental evaluation across several datasets established the superiority of the proposed MI-Hilbert weight initialization approach, which offered a better convergence speed while maintaining training stability when using MSE and R2 metrics for assessment.