Low precision and coarse-to-fine dynamic fixed-point quantization design in convolution neural network

US12014273B2 · kind B2 · utility

Assignee

• Kneron (Taiwan) Co., Ltd. · TW

Inventors

• Jie Wu · Hangzhou City, CN
• Yunhan Ma · San Diego, US
• Bike Xie · San Diego, US
• Hsiang-Tsun Li · San Diego, US
• JUNJIE SU · San Diego, US
• Chun-Chen Liu · Hengshan, TW

Key dates

• Filing date: Aug 27, 2019
• Grant date: Jun 18, 2024
• Expiry date: Aug 16, 2042

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N3/063
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

After inputting input data to a floating pre-trained convolution neural network to generate floating feature maps for each layer of the floating pre-trained CNN model, a statistical analysis on the floating feature maps is performed to generate a dynamic quantization range for each layer of the floating pre-trained CNN model. Based on the obtained quantization range for each layer, the proposed quantization methodologies quantize the floating pre-trained CNN model to generate the scalar factor of each layer and the fractional bit-width of a quantized CNN model. It enables the inference engine performs low-precision fixed-point arithmetic operations to generate a fixed-point inferred CNN model.