Achieving high-fidelity photonic circuits for universal unitaries has been a critical issue for classical and quantum computing applications. The basic strategy for realizing U(n) in photonic systems is to find the algorithm to decompose U(n) into a set of SU(2) operations. While various methods have been implemented for such decomposition, the resulting U(n) may not be optimized for high fidelity, especially when we assume noises in the constituent elements. The programs of this archive describe the analysis of achieving the artificial photonic materials having universal unitary operations, quantifying heavy-tailed distributions in photonic circuits and platforms, examining pruning performance in unitary operations, and applying the pruning to deep neural network applications in order to achieve high-fidelity operations.