Advancing Neural Network Efficiency: From Network Compression Theory To Large Artificial Intelligence Models