FANN: Fourier Adaptive Neural Network for Dynamic Learning

The human brain processes information by dynamically altering the strength of its vast neural connections. In contrast to that conventional artificial neural networks use static weights for prediction which limits its adaptability. To bridge this gap, this study proposes a novel Fourier Adaptive Chebyshev Neural Network. This network emulates biological learning by employing dynamic input-dependent weights. It also integrates a Fourier transformation layer to enrich input features and identify periodic patterns, while Chebyshev polynomials enable the weights to adapt in response to input stimuli. Rigorous benchmarking across diverse real-world and synthetic datasets demonstrated the superior efficacy and generalizability of our approach. The results show the proposed model provides robust and precise classification outcomes.