Validation set feedback method to improve training efficiency and effect of deep network for face recognition
A deep network and face recognition technology, applied in the field of verification set feedback, can solve the problem of lack of closed-loop control in the training process of the deep neural network for face recognition, and achieve the effect of improving the effect and efficiency
Active Publication Date: 2021-05-18
ZHEJIANG UNIV OF TECH
View PDF6 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0004] In order to solve the problem of the lack of closed-loop control in the training process of the existing deep neural network for face recognition, the present invention provides a verification set feedback method for improving the training efficiency and effect of the deep neural network for face recognition. The effect feedback control training process on the verification set is a closed-loop control method
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0023] Below will combine the appended in the embodiment of the present invention figure 1 , clearly and completely describe the technical solutions in the embodiments of the present invention.
[0024] refer to figure 1 , a verification set feedback method for improving the training efficiency and effect of a deep neural network for face recognition, comprising the steps of:
[0025] Step 1: Collect 50,000 ID-labeled face images as a training data set, divide them into non-overlapping training set T (45,000 images) and verification set V (5,000 images) at a ratio of 9:1, select the ResNet network structure, and set Initial learning rate 0.01, SGD learning method, Step=10000, batch_size=100 and other hyperparameters;
[0026] Step 2: Initialize the ResNet model M 0 , evaluate the model M on the validation set V 0 The classification loss R 0 , determine the number N=10 of candidate training schemes;
[0027] Step 3: Randomly scramble the training set T 10 times, and save ...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
A verification set feedback method for improving the training efficiency and effect of a deep neural network for face recognition, comprising the following steps: Step 1: determining a training task, a training data set and an evaluation standard for a deep neural network for face recognition, and dividing the training data set into The ratio is divided into non-overlapping training set and validation set; Step 2: Initialize the model M 0 , to determine the number N of candidate model parameters; Step 3: Randomly sample on the training set to generate N training set sequences; Step 4: Use M 0 As a starting point, use the backpropagation method to update the training model on each training data sequence; Step 5: Evaluate the effect of the N candidate models generated in Step 4 on the verification set; Step 6: Assign each candidate model according to the effect of the model Probability P n , combined with randomness to select a candidate model to replace the model M 0 ;Step 7: Repeat steps 3 to 6 until the training ends. The invention effectively improves the effect and efficiency of deep neural network training.
Description
technical field [0001] The invention relates to the field of machine learning and artificial intelligence algorithms, in particular to a verification set feedback method for improving the training efficiency and effect of a deep neural network for face recognition. Background technique [0002] Deep neural network has greatly improved the performance of machine learning, and has achieved great success in the fields of object detection, pattern recognition, semantic segmentation and natural language processing, and has become a mainstream branch of machine learning theory research and industrial application. How to improve the training efficiency of deep neural networks and improve the effect of training is one of the key issues affecting the development and application of deep neural networks. [0003] The existing face recognition deep neural network training iteratively updates the parameters of the network model through error back propagation, which is an open-loop proces...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/00G06N3/04G06N3/08
CPCG06N3/084G06V40/172G06V40/161G06N3/045
Inventor 高华陈胜勇
Owner ZHEJIANG UNIV OF TECH