Abstract: In recent years the rapid development of intelligent photonics has relied on the collaborative efforts in various cutting-edge directions, including new material structures, new theoretical perspectives, and new device architectures. Optical platforms with excellent nonlinear characteristics have laid the foundation for the rich performance of photonic chips. The introduction of new theories such as non-Hermitian topological photonics has opened up new possibilities for high-performance on-chip processing solutions. New devices for various optical neural network architectures have provided effective pathways for all-optical computation and large-scale integration. Against this backdrop, this review assesses recent research progress in critical areas of intelligent photonics development, including nonlinear materials, non-Hermitian topological photonics, and novel device architectures based on optical neural networks. These developments will have profound implications for the large scale realization of future intelligent photonic chips with higher integration and more complex computational performance, promising broad applications in fields such as all-optical computing, all-optical signal processing, and quantum technologies.