Combined processing device, chip and method for performing determinant inference

Abstract

The invention relates to a combined processing device for executing deterministic reasoning. The combined processing device comprises an instruction extraction module, an instruction pre-decoding module, an instruction decoding module, an instruction execution module, a memory access module and a register write-back module, the instruction extraction module reads the instruction from the instruction memory and updates the program counter to point to the next instruction; the instruction pre-decoding module decodes the compression instruction into a native instruction, and the instruction decoding module accesses the register file and determines branch control; the instruction execution module performs vector calculation or scalar calculation on the instruction to obtain a result, accessesan accessed memory for the Load / Store instruction, calculates branches and skips and performs inspection according to an expected result; in the memory access module, memory access of the assembly line is completed; and the register write-back module writes the result of the execution stage into the register file. The invention further relates to an operation device of the deterministic neural network algorithm comprising the combined processing device, a computing chip and application of the computing chip.

Description

technical field [0001] The application relates to a combined processing device, chip and method for performing deterministic reasoning, which are suitable for the field of computer technology. Background technique [0002] Artificial neural network, referred to as neural network, in the field of machine learning and cognitive science, is a mathematical model or computational model that imitates the structure and function of biological neural networks, and is used to estimate and approximate functions. A neural network is composed of a large number of nodes (or neurons) connected to each other. Each connection between two nodes represents a weighted value for the signal passing through the connection, called weight, which is equivalent to the memory of the artificial neural network. The output of the network varies according to the way the network is connected, the weight value and the activation function. The network itself is usually an approximation to a certain algorith...

AI Technical Summary

Problems solved by technology

Second, existing algorithms do not have the function of security protection for training data and inference results

Affected by this, the application of neural network algorithms in fields with higher security requirements such as finance, trusted computing, blockchain, and smart contracts has been greatly restricted.

Images