Nvidia corporation (20240112007). NEURAL NETWORK ACCELERATOR USING LOGARITHMIC-BASED ARITHMETIC simplified abstract

NEURAL NETWORK ACCELERATOR USING LOGARITHMIC-BASED ARITHMETIC

Organization Name

nvidia corporation

Inventor(s)

William James Dally of Incline Village CA (US)

Rangharajan Venkatesan of San Jose CA (US)

Brucek Kurdo Khailany of Austin TX (US)

NEURAL NETWORK ACCELERATOR USING LOGARITHMIC-BASED ARITHMETIC - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240112007 titled 'NEURAL NETWORK ACCELERATOR USING LOGARITHMIC-BASED ARITHMETIC

Simplified Explanation

Neural networks often include convolution layers that perform multiplication and addition operations. Multiplication on logarithmic format values is energy efficient as exponents are simply added, but addition is more complex. A new method involves decomposing exponents into quotient and remainder components, sorting based on remainders, summing quotients, multiplying by remainders, and converting back to logarithmic format.

• Logarithmic format values can be added by decomposing exponents into quotient and remainder components.
• Quotient components are sorted based on remainder components.
• Sorted quotient components are summed to produce partial sums.
• Partial sums are multiplied by remainder components to produce a final sum.
• The final sum is converted back into the logarithmic format.

Potential Applications

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

Problems Solved

This innovation solves the complexity and energy inefficiency issues associated with performing addition on logarithmic format values in neural networks.

Benefits

The benefits of this technology include improved energy efficiency, faster computation, and enhanced performance in neural network applications.

Potential Commercial Applications

This technology can be commercialized in industries such as healthcare for medical image analysis, finance for fraud detection, and autonomous vehicles for object recognition.

Possible Prior Art

One possible prior art could be the use of fixed-point or floating-point formats for performing multiplication and addition operations in neural networks.

How does this method compare to existing techniques for adding logarithmic format values in terms of computational efficiency?

This method is more computationally efficient as it avoids the need to convert values to integers and back to logarithmic format, reducing the overall computational complexity.

What impact does the use of logarithmic format values have on the accuracy of neural network models?

The use of logarithmic format values may have minimal impact on the accuracy of neural network models, as the focus is on improving energy efficiency and computational speed rather than altering the underlying model architecture.

Original Abstract Submitted

neural networks, in many cases, include convolution layers that are configured to perform many convolution operations that require multiplication and addition operations. compared with performing multiplication on integer, fixed-point, or floating-point format values, performing multiplication on logarithmic format values is straightforward and energy efficient as the exponents are simply added. however, performing addition on logarithmic format values is more complex. conventionally, addition is performed by converting the logarithmic format values to integers, computing the sum, and then converting the sum back into the logarithmic format. instead, logarithmic format values may be added by decomposing the exponents into separate quotient and remainder components, sorting the quotient components based on the remainder components, summing the sorted quotient components to produce partial sums, and multiplying the partial sums by the remainder components to produce a sum. the sum may then be converted back into the logarithmic format.