Quantizing autoencoders in a neural network
AssigneeNVIDIA Corporation
InventorsJon Hasselgren, Jacob Munkberg
Life of this patent
Estimating the term…
Abstract
The performance of a neural network is improved by applying quantization to data at various points in the network. In an embodiment, a neural network includes two paths. A quantization is applied to each path, such that when an output from each path is combined, further quantization is not required. In an embodiment, the neural network is an autoencoder that includes at least one skip connection. In an embodiment, the system determines a set of quantization parameters based on the characteristics of the data in the primary path and in the skip connection, such that both network paths produce output data in the same fixed point format. As a result, the data from both network paths can be combined without requiring an additional quantization.
Claims20
A computer-implemented method, comprising:
- calculating one or more first and second quantization parameters corresponding to a respective first and second portion of a neural network;
- and
- applying at least a portion of the first and second quantization parameters to the first and second portions of the neural network, wherein the first and second quantization parameters are selected to obviate the need to further quantize the output of the first and second portions of the neural network.
A system, comprising:
- one or more processors;
- and
- memory to store computer-executable instructions that, if executed, cause the system to:
- determine a set of quantization parameters for first data in a first path of a neural network and second data in a second path of the neural network, such that a first output of the first path and a second output of the second path can be combined without further quantization;
- quantize the first data and the second data in accordance with the set of quantization parameters;
- and
- combine the first output and the second output.
A non-transitory computer-readable storage medium having stored thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to at least:
- determine a set of quantization parameters for first data in a first path of a neural network and second data in a second path of the neural network, such that a first output of the first path and a second output of the second path can be combined without further quantization;
- quantize the first data and the second data in accordance with the set of quantization parameters;
- and
- combine the first output and the second output.
Dependencies are inferred from the claim text.