17887216. QUANTIZATION METHOD OF NEURAL NETWORK AND APPARATUS FOR PERFORMING THE SAME simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

QUANTIZATION METHOD OF NEURAL NETWORK AND APPARATUS FOR PERFORMING THE SAME

Organization Name

SAMSUNG ELECTRONICS CO., LTD.

Inventor(s)

Jun-Woo Jang of Suwon-si (KR)

Jaewoo Park of Ulsan (KR)

Faaiz Asim of Ulsan (KR)

Jongeun Lee of Ulsan (KR)

QUANTIZATION METHOD OF NEURAL NETWORK AND APPARATUS FOR PERFORMING THE SAME - A simplified explanation of the abstract

This abstract first appeared for US patent application 17887216 titled 'QUANTIZATION METHOD OF NEURAL NETWORK AND APPARATUS FOR PERFORMING THE SAME

Simplified Explanation

The abstract describes a method and apparatus for quantizing the parameters of a neural network. The method involves obtaining the parameters of the neural network and quantizing them using a quantization scheme that excludes zero and includes positive and negative quantization levels symmetric to each other. The quantized parameters are then outputted.

• The method quantizes the parameters of a neural network.
• It uses a quantization scheme that excludes zero from quantization levels.
• The quantization scheme includes positive and negative quantization levels symmetric to each other.
• The quantized parameters are outputted.

Potential Applications

• This technology can be applied in various fields that utilize neural networks, such as machine learning, artificial intelligence, and data analysis.
• It can be used in image recognition systems, natural language processing, and speech recognition applications.
• The quantization method can be implemented in hardware or software for efficient processing of neural networks.

Problems Solved

• Neural networks often require a large number of parameters, which can lead to high computational and memory requirements.
• Quantization helps reduce the memory footprint and computational complexity of neural networks.
• The method solves the problem of quantizing neural network parameters while maintaining accuracy and performance.

Benefits

• The quantization method reduces the memory requirements of neural networks, allowing for more efficient storage and processing.
• It also reduces the computational complexity, enabling faster inference and training of neural networks.
• The method provides a balance between accuracy and efficiency, making it suitable for various applications.

Original Abstract Submitted

A quantization method of a neural network, and an apparatus for performing the quantization method are provided. The quantization method includes obtaining parameters of the neural network, quantizing the parameters using a quantization scheme in which at least one positive quantization level and at least one negative quantization level symmetric to each other by excluding zero from quantization levels, and outputting the quantized parameters.