Depth symbol regression accelerator and depth symbol regression method

The invention provides a depth symbol regression accelerator comprising an execution unit array used for NN inference operation and BFGS optimization operation of depth symbol regression; the on-chip cache is in communication connection with the execution unit array and comprises an input and output cache used for storing input and output values of the NN inference operation and the BFGS optimization operation; the NN weight cache is used for storing the weight of the NN inference model; the parameter cache is used for storing configuration parameters of the execution unit array; the character string analysis module is in communication connection with the input/output cache and is used for converting character strings of a symbolized expression framework obtained by NN reasoning operation into operators of a mathematical expression; and the controller is in communication connection with the execution unit array and the on-chip cache and is used for controlling the memory access operation and operation schedul