8-bit integer full-quantization reasoning method and device based on self-adaptive dynamic shifting
A reasoning method and adaptive technology, applied in the field of convolutional neural networks, can solve problems such as unavailability, and achieve the effect of avoiding overflow errors, reducing errors, and reducing on-chip memory
Pending Publication Date: 2020-11-17
BEIJING UNISOUND INFORMATION TECH +1
View PDF3 Cites 3 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0008] Third, the existing technology still needs to perform floating-point calculations in reasoning, which is not available for artificial intelligence chips such as AI Chips that only support fixed-point calculations.
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
[0057] The application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain related inventions, rather than to limit the invention. It should also be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.
[0058] It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.
[0059] In order to facilitate the understanding of the present invention, some of the professional terms involved are firstly explained:
[0060] Split is similar to copying, and Cube represents the copied block;
[0061] Conv is an ordinary convolution;
[0062] DwConv is...
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
The invention provides an 8-bit integer full-quantization reasoning method and device based on self-adaptive dynamic shifting. The method comprises the steps of acquiring a trained floating point model; obtaining the weight of each channel in a floating point model; calculating an activation value of each layer in the floating point model through KLD; on the basis of the activation value, determining a conversion factor for a skip layer and convolution channel disruption operation of the floating point model, and pre-storing all fixed-point values and shift values; and obtaining the weight scale of the fixed point of the floating point model according to a quantization table, and outputting an integer result based on the weight. According to the method provided by the embodiment of the invention, errors of conversion from floating points to fixed points are greatly reduced according to channel full-fixed-point quantification, floating point operation and full-fixed-point shifting operation are not involved in the reasoning process, whether result errors meet the requirements of an artificial intelligence chip or not after full quantification of the model can be verified, and self-adaptive dynamic shifting is realized; an overflow error caused by fixed displacement is avoided, and an int32 intermediate value is optimized into an int8 intermediate value, so that an on-chip memoryis further reduced.
Description
technical field [0001] One or more embodiments of the present invention relate to the technical field of Convolutional Neural Networks (CNNs), in particular to an 8-bit integer full quantization reasoning method and device based on adaptive dynamic shifting. Background technique [0002] This section is intended to provide a background or context for implementations of the invention that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section. [0003] CNNs have achieved superior results in the fields of image classification, object detection, face recognition, etc. However, due to the complexity of the network structure and calculation delay, the real-time forward reasoning of CNNs can be realized on embedded platforms with relatively insufficient storage resources and computing resources. , it is necessary to compress the model size of the neural network and improve the computational efficiency of the model under th...
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 Applications(China)
IPC IPC(8): G06N3/04G06N3/063G06N3/08
CPCG06N3/063G06N3/08G06N3/048G06N3/045
Inventor 谢远东
Owner BEIJING UNISOUND INFORMATION TECH