Neural network model optimization method based on class expansion learning

Abstract

The invention discloses a neural network model optimization method based on class expansion learning. The method specifically comprises the following steps: obtaining a data set which is used for training and contains various types of samples, and defining an algorithm target; extracting features of each type of pictures in the data set by using a universal model, and evaluating the error-prone degree of each type according to the distribution condition of each type of features; adding several types of data with the highest error-prone degree into a training pool, and optimizing the neural network by using the data in the training pool; after the optimization is completed, adding the remaining data with the highest error-prone degree into a training pool, expanding the categories in the training pool, and further optimizing the neural network by using the training pool on the basis of the neural network obtained by the last training; and continuously performing class expansion on the training pool until the whole data set enters the training pool to obtain a final optimized neural network model. The method is suitable for neural network model optimization based on multi-class datasets in supervised learning, and has good effect and robustness for various complex conditions.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to a neural network model optimization method based on class expansion learning. Background technique [0002] The optimization method of the neural network model is the underlying technology of artificial intelligence, and it is often used as the basis for high-level visual tasks, such as object detection, target recognition, semantic segmentation, etc. However, limited by computer computing resources and memory resources, current optimization methods for neural network models rely on batch stochastic gradient descent. This method is an iterative, batch-level learning model. Each training cannot use global data, but only one batch of data. Since the data for each training is usually distributed in an extremely sparse and scattered space, the optimization of the neural network model is very difficult, and it will be affected by most of the simple data while optimizing, ignoring the i...

AI Technical Summary

Problems solved by technology

However, this type of optimization method has many details, high implementation cost, and difficult to reproduce; on the other hand, this type of method often designs specific evaluation indicators for specific tasks, and its evaluation indicators do not have the ability to migrate

Images