Method and apparatus with neural network control
AssigneesSamsung Electronics Co., Ltd., Seoul National University R&Db Foundation
InventorsJongseok Kim, Hyun Oh Song, Yeonwoo Jeong
Life of this patent
Estimating the term…
Abstract
A method and apparatus with neural network control are provided. In one general aspect, a method is performed by at least one processor of an apparatus, the method includes selecting neural networks from among available neural networks with respective input image resolutions, including plural resolutions, wherein the selected neural networks are selected based on a sum of amounts of mutual information between the selected neural networks being minimal and based on a limiting condition of a use of the available neural networks, and configuring the available neural networks to perform an inference on input data, wherein the configuring is based on the combination of the selected neural networks.
Claims23
A method performed by one or more processors of an apparatus, the method comprising:
- selecting neural networks from among available neural networks with respective input image resolutions, including plural resolutions, wherein the selected neural networks are selected based on a sum of amounts of mutual information between the selected neural networks being minimal and based on a limiting condition of a use of the available neural networks, wherein the amount of mutual information of the combination is minimum among amounts of mutual information of respective combinations of the available neural networks, and wherein the combinations are formed based on a limiting condition of an intended use of the available neural networks;
- and
- configuring the available neural networks to perform an inference on input data, wherein the configuring is based on the combination of the selected neural networks.
A method performed by one or more processors of an apparatus, the method comprising:
- accessing, among available neural networks with respective input image resolutions, including plural resolutions, a combination of the available neural networks, wherein neural networks in the combination are selected based on an amount of mutual information therebetween, and wherein the amount of mutual information of the combination is minimum among amounts of mutual information of respective combinations of the available neural networks, wherein the combinations are formed based on a limiting condition of an intended use of the available neural networks;
- and
- configuring the neural networks in the combination to perform an inference on input data based on an inference-time limiting condition.
An apparatus comprising:
- one or more processors;
- a memory storing instructions configured to cause the one or more processors to:
- select neural networks from among available neural networks, wherein the available neural networks are configured for respective input image resolutions, including plural image resolutions, wherein the selected neural networks are selected based on a sum of amounts of mutual information between the selected neural networks being minimal and based on a limiting condition of a use of the available neural networks, wherein the amount of mutual information of the combination is minimum among amounts of mutual information of respective combinations of the available neural networks, and wherein the combinations are formed based on a limiting condition of an intended use of the available neural networks;
- and
- configure the available neural networks to perform an inference on input data, wherein the configuring is based on the combination of the selected neural networks.
An apparatus comprising:
- one or more processors;
- a memory storing instructions configured to cause the one or more processors to:
- acquire available neural networks each trained for a different respective input image resolution, wherein a combination of the available neural networks is selected based on an amount of mutual information between at least a pair of the available neural networks, wherein the amount of mutual information of the combination is minimum among amounts of mutual information of respective combinations of the available neural networks, and wherein the combinations are formed based on a limiting condition of an intended use of the available neural networks;
- and
- configure a neural network in the selected combination to perform inference on input data, based on a limiting condition.
Dependencies are inferred from the claim text.