Qualcomm incorporated (20240104356). QUANTIZED NEURAL NETWORK ARCHITECTURE simplified abstract

QUANTIZED NEURAL NETWORK ARCHITECTURE

Organization Name

qualcomm incorporated

Inventor(s)

Srijesh Sudarsanan of Waltham MA (US)

Deepak Mathew of Acton MA (US)

Marc Hoffman of Mansfield MA (US)

Sundar Rajan Balasubramanian of Groton MA (US)

Gerald Sweeney of Chelmsford MA (US)

Mansi Jain of Littleton MA (US)

James Lee of Northborough MA (US)

Ankita Nayak of Milpitas CA (US)

QUANTIZED NEURAL NETWORK ARCHITECTURE - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240104356 titled 'QUANTIZED NEURAL NETWORK ARCHITECTURE

Simplified Explanation

The present disclosure relates to techniques for quantized machine learning, specifically in the context of neural networks.

• A quantized input matrix is accessed at a layer of a neural network.
• A first interim value is generated in an accumulator by performing matrix multiplication of the quantized input matrix and a quantized weight matrix associated with the layer.
• The first interim value is normalized based on leading sign bits and then dequantized.
• A second interim value is generated by applying a rounded right-shift operation to the dequantized normalized first interim value.
• Activation data is generated by applying an activation function to the second interim value.

Potential Applications

This technology can be applied in various fields such as image recognition, natural language processing, and autonomous driving.

Problems Solved

This technology helps in reducing the computational complexity and memory requirements of neural networks, making them more efficient and faster.

Benefits

The benefits of this technology include improved performance, reduced power consumption, and faster inference times in neural network applications.

Potential Commercial Applications

Potential commercial applications of this technology include smart devices, healthcare diagnostics, and financial analysis.

Possible Prior Art

Prior art in the field of quantized machine learning includes research papers on low-bit quantization techniques and optimization algorithms for neural networks.

How does this technology compare to traditional machine learning methods in terms of accuracy and efficiency?

This article does not provide a direct comparison between quantized machine learning and traditional methods in terms of accuracy and efficiency. Further research or experimentation may be needed to address this question.

What are the limitations of this technology in real-world applications?

The article does not discuss the potential limitations of implementing quantized machine learning in real-world applications. Understanding these limitations could help in better assessing the practicality of this technology.

Original Abstract Submitted

certain aspects of the present disclosure provide techniques and apparatus for quantized machine learning. a quantized input matrix is accessed at a layer of a neural network, and a first interim value is generated in an accumulator by performing matrix multiplication, using the accumulator, of the quantized input matrix and a quantized weight matrix associated with the layer of the neural network. the first interim value is normalized based at least in part on one or more leading sign bits of the first interim value, and the normalized first interim value is dequantized. a second interim value is generated by applying a rounded right-shift operation to the dequantized normalized first interim value, and activation data is generated by applying an activation function to the second interim value.