Deep neural network quantification method, system and device, and medium
A deep neural network and neural network technology, applied in neural learning methods, biological neural network models, inference methods, etc., can solve problems such as the reduction of inference accuracy, and achieve the effects of high efficiency, simple model evaluation criteria, and strong universality.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0048] Please refer to figure 1 , figure 1 It is a schematic flow chart of a deep neural network quantification method. Embodiment 1 of the present invention provides a deep neural network quantification method, the method comprising:
[0049] Get the first deep neural network , the Including n neural network layers, the neural network layers are divided into quantized layers and non-quantized layers, the The accuracy rate of , set the highest acceptable accuracy loss threshold for the quantized deep neural network ;
[0050] Based on the and said , using a dichotomy to search for all quantized layers from the n neural network layers, and quantize the obtained quantized layers.
[0051] The following is a detailed introduction to this method in combination with specific examples and existing deep neural network optimization methods:
[0052] This embodiment first introduces the prior art relevant to the present invention, and purpose is to highlight the differe...
Embodiment 2
[0131] Please refer to figure 2 , figure 2 It is a schematic diagram of the composition of a deep neural network quantization system. Embodiment 2 of the present invention provides a deep neural network quantization system, and the system includes:
[0132] Network accuracy rate and accuracy rate loss threshold acquisition unit, used to obtain the first deep neural network , the Including n neural network layers, the neural network layers are divided into quantized layers and non-quantized layers, the The accuracy rate of , set the highest acceptable accuracy loss threshold for the quantized deep neural network ;
[0133] quantization unit for the and said , using a dichotomy to search for all quantized layers from the n neural network layers, and quantize the obtained quantized layers.
Embodiment 3
[0135] Embodiment 3 of the present invention provides a deep neural network quantification device, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, Steps for realizing the quantization method of the deep neural network.
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More