Method and device for determining performance parameter value of network, equipment and medium

Abstract

The embodiment of the invention provides a method and device for determining a performance parameter value of a network, equipment and a medium, and belongs to the technical field of machine learning, and the method comprises the steps of enabling a plurality of network modules to be combined with an original network, and obtaining a plurality of subnetworks of the original network, wherein one network module is used for realizing one type of data processing operation; alternately training the plurality of sub-networks for a plurality of times to complete the training of the original network, the sub-networks including the same network module sharing the parameters of the network module; and for any sub-network in the plurality of sub-networks, determining a performance parameter value of the sub-network according to the accumulated gradient value of each network module included in the sub-network in the training process of the original network.

Description

technical field [0001] The present application relates to the technical field of machine learning, in particular to a method, device, device and medium for determining network performance parameter values. Background technique [0002] With the development of deep learning, the performance of deep neural networks has achieved significant performance improvements in various visual tasks and speech tasks, but the existing neural network structures, such as ResNet, DenseNet, etc., only propose a general model Topological structure paradigms, although these paradigms can achieve better results in some general computer tasks, but with the complexity of application scenarios, it is necessary to target different tasks and different conditional constraints (such as computing power constraints in embedded devices) , storage constraints, etc.) to adapt the model structure to achieve the best balance between performance and conditional constraints. However, this process requires a hig...

AI Technical Summary

Problems solved by technology

However, since the accuracy of the sub-model on the verification data set is not enough to reflect the real performance of the sub-model, the performance of the subsequent searched sub-model in the inference stage is not high

Images