Algorithm verification method and system for federated learning heterogeneous processing system

Abstract

The embodiment of the invention provides an Algorithm verification method and system for federated learning heterogeneous processing system, which is applied to verifying the correctness of an FPGA heterogeneous computing chip algorithm in the heterogeneous processing system, and comprises the following steps: acquiring configuration information which comprises an algorithm type, each input parameter and the bit width of each input parameter; randomly generating input parameter data according to the bit width configuration information of each input parameter; performing calculation based on astandard operation library according to the input parameter data to obtain a standard value; based on an actual algorithm of an FPGA heterogeneous computing chip, calculating according to the input parameter data to obtain an actual value; and comparing the standard value with the actual value, if the standard value is consistent with the actual value, determining that the algorithm verification is passed, and if the standard value is inconsistent with the actual value, sending error information. According to the scheme, the correctness of algorithm implementation on the heterogeneous computing chip can be automatically verified, and the engineering verification efficiency is improved.

Description

technical field [0001] The embodiments of this specification generally relate to the field of heterogeneous computing, and in particular to an algorithm verification method and system for federated learning heterogeneous processing systems. Background technique [0002] The present invention relates to a federated learning heterogeneous system in the field of artificial intelligence (AI). Federated learning is a distributed machine learning framework. Under the premise of ensuring data security, privacy, and legality, different participants can achieve common modeling. To share data securely and privately between different participants, the data must first be encrypted homomorphically. Homomorphic encryption is a mathematical operation with high bit width and large integers. The amount of calculation is very complicated, and it also involves real-time calculation, etc It is required that in practical applications, the realization of homomorphic encryption gradually changes ...

AI Technical Summary

Problems solved by technology

[0004] The hardware-based algorithm verification solution requires designing circuits, or converting the algorithm into machine language, compiling, laying out and wiring, and downloading to heterogeneous chips, and the hardware design cycle is very long, often weeks or even months, and each algorithm change , have to change the hardware design, resulting in a very long iterative cycle of verification

Images