Neural Network Representation Standard Framework Structure

Abstract

The present invention provides a neural network representation standard framework structure, including: an interoperable representation module, which converts the input neural network to obtain an interoperable representation format, which includes the syntax definition of the neural network and the supported operation definition and weight format definition; the compact representation module converts the interoperable neural network into a serialized format through the neural network compression algorithm, which includes the syntax definition of the compressed neural network, the supported operation definition and the weight format definition; The decoding representation module converts the compact representation of the neural network into a codec representation through the neural network compression algorithm, which includes the syntax definition of the compressed neural network, the definition of supported operation operations and the definition of the weight format after codec; the package representation module will Security information and authentication information are encapsulated with the neural network, thereby converting the neural network into a model.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a neural network representation standard framework structure in deep learning. Background technique [0002] As the core driving force of a new round of industrial transformation, artificial intelligence will reconstruct all aspects of economic activities such as production, distribution, exchange, and consumption, form new demands for intelligence in various fields from macro to micro, and give birth to new technologies, new products, and new technologies. Industry, new format, new model. Deep learning is the core technology of artificial intelligence development in recent years, and neural network representation is the basic problem in the application of deep learning technology. Among them, the convolutional neural network is currently the most widely used deep neural network. It is driven by a large number of application fields such as handwritten digit recognition, fa...

AI Technical Summary

Problems solved by technology

However, today's compression algorithms all rely on different deep learning open source algorithm frameworks, resulting in the incompatibility of compressed neural network models generated on different deep learning algorithm frameworks, and there is no unified representation of these compressed neural networks , leading to poor compatibility of the model, hindering the application and development of deep learning algorithms on restricted devices

Images