Apparatus, system, and method of generating a multi-model Machine Learning (ML) architecture
For example, an apparatus may include an input to receive Machine Learning (ML) model information corresponding to an ML model to process input information; and a processor to construct a multi-model ML architecture including a plurality of ML model variants based on the ML model, wherein the processor is configured to determine the plurality of ML model variants based on an attribution-based diversity metric corresponding to a model group including a first ML model variant and a second ML model variant, wherein the attribution-based diversity metric corresponding to the model group is based on a diversity between a first attribution scheme and a second attribution scheme, the first attribution scheme representing first portions of the input information attributing to an output of the first ML model variant, the second attribution scheme representing second portions of the input information attributing to an output of the second ML model variant.
1 . An apparatus comprising:
an input to receive Machine Learning (ML) model information corresponding to an ML model to process input information; and
a processor configured to construct a multi-model ML architecture comprising a plurality of ML model variants based on the ML model, wherein the processor is configured to determine a plurality of derived ML models based on the ML model, and to select the plurality of ML model variants from the plurality of derived ML models based on an attribution-based diversity metric corresponding to a model group comprising a first ML model variant and a second ML model variant, wherein the attribution-based diversity metric corresponding to the model group is based on a diversity between a first attribution scheme and a second attribution scheme, the first attribution scheme representing one or more first portions of the input information attributing to an output of the first ML model variant, the second attribution scheme representing one or more second portions of the input information attributing to an output of the second ML model variant.
2 . The apparatus of claim 1 , wherein the processor is configured to select a plurality of ML model candidates from the plurality of derived ML models, and to select the plurality of ML model variants from the plurality of ML model candidates based on the attribution-based diversity metric.
3 . The apparatus of claim 2 , wherein the processor is configured to select the plurality of ML model candidates from the plurality of derived ML models based on a performance criterion corresponding to performance of the plurality of derived ML models.
4 . The apparatus of claim 1 , wherein the processor is configured to determine a plurality of attribution-based diversity metric scores corresponding to a plurality of model groups, and to select the plurality of ML model variants from the plurality of derived ML models based on the plurality of attribution-based diversity metric scores.
5 . The apparatus of claim 4 , wherein the processor is configured to determine a plurality of performance scores corresponding to the plurality of model groups, and to select the plurality of ML model variants from the plurality of derived ML models based on the plurality of performance scores.
6 . The apparatus of claim 1 , wherein the processor is configured to select the plurality of ML model variants from the plurality of derived ML models based on a performance criterion corresponding to performance of the plurality of derived ML models.
7 . The apparatus of claim 1 , wherein the processor is configured to generate the plurality of derived ML models based on a Neural Architecture Search (NAS).
8 . The apparatus of claim 1 , wherein the processor is configured to generate the plurality of ML model variants based on the attribution-based diversity metric.
9 . The apparatus of claim 1 , wherein the processor is configured to determine the multi-model ML architecture based on the attribution-based diversity metric.
10 . The apparatus of claim 1 , wherein the processor is configured to select the multi-model ML architecture from a plurality of multi-model ML architectures based on computing resources of a computing device to execute one or more of the plurality of ML model variants.
11 . The apparatus of claim 1 , wherein the processor is configured to determine a first attribution-based diversity metric score corresponding to a first group of ML models and to determine a second attribution-based diversity metric score corresponding to a second group of ML models, wherein the first attribution-based diversity metric score is higher than the second attribution-based diversity metric score, and wherein a diversity between attribution schemes of ML models in the first group of ML models is greater than a diversity between attribution schemes of ML models in the second group of ML models.
12 . The apparatus of claim 1 , wherein the ML model comprises an image classification ML model to process image information of an image, and wherein the first attribution scheme represents a level of attribution of one or more pixels of the image to an output of the first ML model variant, and the second attribution scheme represents a level of attribution of one or more second pixels of the image to an output of the second ML model variant.
13 . The apparatus of claim 1 , wherein the ML model comprises an ML model to process the input information at an autonomous robot.
14 . The apparatus of claim 1 , wherein the processor is configured to determine the multi-model ML architecture comprising a first plurality of ML model variants to be executed on a first computing device, and a second plurality of ML model variants to be executed on a second computing device.
15 . The apparatus of claim 14 , wherein the processor is configured to determine the first plurality of ML model variants based on computing resources of the first computing device, and to determine the second plurality of ML model variants based on computing resources of the second computing device.
16 . The apparatus of claim 14 , wherein the processor is configured to execute the first plurality of ML model variants, and to provide the second plurality of ML model variants to the second computing device.
17 . The apparatus of claim 16 , wherein the first plurality of ML model variants is configured to perform one or more first parts of a task, and the second plurality of ML model variants is configured to perform one or more second parts of the task.
18 . The apparatus of claim 17 , wherein the one or more second parts of the task comprise fail-safe operations corresponding to one or more fail-safe events of the task.
19 . The apparatus of claim 17 , wherein the one or more second parts of the task comprise safety operations corresponding to one or more safety events.
20 . The apparatus of claim 14 , wherein the first computing device comprises a server, and the second computing device comprises an autonomous robot.
21 . A product comprising one or more tangible computer-readable non-transitory storage media comprising computer-executable instructions operable to, when executed by at least one processor, enable the at least one processor to cause a computing device to:
process Machine Learning (ML) model information corresponding to an ML model to process input information; and
construct a multi-model ML architecture comprising a plurality of ML model variants based on the ML model by determining the plurality of ML model variants based on an attribution-based diversity metric corresponding to a model group comprising a first ML model variant and a second ML model variant, wherein the attribution-based diversity metric corresponding to the model group is based on a diversity between a first attribution scheme and a second attribution scheme, the first attribution scheme representing one or more first portions of the input information attributing to an output of the first ML model variant, the second attribution scheme representing one or more second portions of the input information attributing to an output of the second ML model variant, wherein the instructions, when executed, cause the computing device to determine a first attribution-based diversity metric score corresponding to a first group of ML models and to determine a second attribution-based diversity metric score corresponding to a second group of ML models, wherein the first attribution-based diversity metric score is higher than the second attribution-based diversity metric score, and wherein a diversity between attribution schemes of ML models in the first group of ML models is greater than a diversity between attribution schemes of ML models in the second group of ML models.
22 . The product of claim 21 , wherein the instructions, when executed, cause the computing device to determine a plurality of derived ML models based on the ML model, and to select the plurality of ML model variants from the plurality of derived ML models based on the attribution-based diversity metric.
23 . An apparatus comprising:
means for inputting Machine Learning (ML) model information corresponding to an ML model to process input information; and
means for constructing a multi-model ML architecture comprising a plurality of ML model variants based on the ML model by determining the plurality of ML model variants based on an attribution-based diversity metric corresponding to a model group comprising a first ML model variant and a second ML model variant, wherein the attribution-based diversity metric corresponding to the model group is based on a diversity between a first attribution scheme and a second attribution scheme, the first attribution scheme representing one or more first portions of the input information attributing to an output of the first ML model variant, the second attribution scheme representing one or more second portions of the input information attributing to an output of the second ML model variant, wherein the means for constructing the multi-model ML architecture comprises means for determining the multi-model ML architecture comprising a first plurality of ML model variants to be executed on a first computing device, and a second plurality of ML model variants to be executed on a second computing device.
24 . The apparatus of claim 23 comprising means for determining a first attribution-based diversity metric score corresponding to a first group of ML models, and determining a second attribution-based diversity metric score corresponding to a second group of ML models, wherein the first attribution-based diversity metric score is higher than the second attribution-based diversity metric score, and wherein a diversity between attribution schemes of ML models in the first group of ML models is greater than a diversity between attribution schemes of ML models in the second group of ML models.