Optimization method and device for neural network

Abstract

The embodiments of the invention disclose an optimization method and device for a neural network. The method comprises the steps of: acquiring a first neural network satisfying a set precision condition, and processing a training sample set based on the first neural network to obtain a first feature vector of each training sample in the training sample set; constructing second neural networks to be trained based on a set network construction condition; training the second neural networks according to the first feature vectors and the training sample set, and determining a second neural network satisfying the set precision condition; and determining the second neural network as a target neural network of the first neural network. By using the method, another newly-constructed small-scale neural network can be directly trained and learnt according to the optimization condition and then determined as a target optimization network of the neural network to be optimized, so that when feature recognition is performed based on the optimized neural network, the purposes of improving the recognition speed, shortening the recognition time and reducing the spatial occupation of a memory, a running memory, a display memory and the like can be fulfilled.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of artificial neural networks, and in particular to a neural network optimization method and device. Background technique [0002] At present, face recognition is usually performed based on a trained neural network model (such as a deep convolutional neural network model). When using the neural network model to carry out face recognition, the existing problems are as follows: 1. The complexity of the image data processing calculation is high, which affects the operation time (such as processing the face image on an electronic device equipped with a Core i7 processor It often takes more than 1 second); 2. It needs to take up a lot of memory space or video memory space of the graphics card during the processing; 3. It also needs to take up a lot of storage space to store the entire neural network model. [0003] The existing neural network model optimization methods cannot completely sol...

AI Technical Summary

Problems solved by technology

When using the neural network model to carry out face recognition, the existing problems are as follows: 1. The complexity of image data processing and calculation is high, which affects the operation time (such as processing face images on electronic devices equipped with Core i7 processors It often takes more than 1 second); 2. It needs to take up a lot of memory space or graphics card memory space during the processing; 3. It also needs to take up a lot of storage space to store the entire neural network model

[0003] The existing neural network model optimization methods cannot completely solve the above-mentioned problems. For example, the optimization in the form of Huffman coding can ensure the processing and calculation accuracy of the optimized neural network model and effectively reduce the depth of the neural network. The storage space of the network model, but it cannot reduce the complexity of processing operations, shorten the running time, and at the same time, it cannot reduce the space occupied by memory or video memory during processing.

Images