Method and system for bit quantization of artificial neural network
AssigneeDeepx Co., Ltd.
InventorLok Won KIM
Life of this patent
Estimating the term…
Abstract
The present disclosure provides a method for bit quantization of an artificial neural network. This method may comprise: (a) a step of selecting one parameter or one parameter group to be quantized in an artificial neural network; (b) a bit quantization step of reducing the size of data representation for the selected parameter or parameter group to bits; (c) a step of determining whether the accuracy of the artificial neural network is greater than or equal to a predetermined target value; and (d) a step of, when the accuracy of the artificial neural network is greater than or equal to the target value, repeatedly performing said step (a) to step (c).
Claims21
A method for quantizing bits of an artificial neural network of a multi-layer structure, the method comprising:
- selecting at least one layer in descending order of computation amount, among a plurality of layers of the artificial neural network of the multi-layer structure;
- bit quantizing to reduce a size of a data representation for a parameter of the selected at least one layer to a unit of bits;
- after the bit quantization, determining whether an accuracy of the artificial neural network of the multi-layer structure is greater than or equal to a target value;
- and
- repeatedly executing the bit quantizing when the accuracy of the artificial neural network of the multi-layer structure is greater than or equal to the target value.
-41. (canceled)
A method for quantizing bits of an artificial neural network of a multi-layer structure, the method comprising:
- selecting at least one parameter for which a search of accuracy fluctuation point is not performed, wherein a bit size of data for a plurality of parameters including at least one of weight data, feature map data, or activation map data of the artificial neural network of the multi-layer structure is fixed as a maximum;
- bit quantizing to reduce a size of a data representation for the selected at least one parameter to a unit of bits;
- after the bit quantization, determining whether an accuracy of the artificial neural network of the multi-layer structure is greater than or equal to a target value;
- and
- repeatedly executing the bit quantizing when the accuracy of the artificial neural network of the multi-layer structure is greater than or equal to the target value.
A method for quantizing bits of an artificial neural network of a multi-layer structure, the method comprising:
- selecting a parameter group based on a computation amount or a memory amount among a plurality of parameters of the artificial neural network of the multi-layer structure;
- bit quantizing to reduce a size of a data representation for the selected parameter group to a unit of bits;
- after the bit quantization, determining whether an accuracy of the artificial neural network of the multi-layer structure is greater than or equal to a target value;
- and
- repeatedly executing the bit quantizing when the accuracy of the artificial neural network of the multi-layer structure is greater than or equal to the target value.
Dependencies are inferred from the claim text.