Synthesis of Parallel-Pipeline Devices for Vertical Computation of Basic Multi-Operand Neural Operations
This paper presents a study of hardware implementation of basic multi-operand neural ope- rations for artificial neural networks. The operational basis of neural networks is identified, which includes groups of preprocessing operations, processor operations, and transfer function computations. The selection of basic multi-operand neural operations is justified: finding extreme values in one- dimensional arrays, calculating the sum of squared differences, scalar product computation, and group summation.